Skip to content

VirtualDB

VirtualDB provides a SQL query interface across heterogeneous HuggingFace datasets using an in-memory DuckDB database. Each dataset defines experimental conditions in its own way, with properties stored at different hierarchy levels (repository, dataset, or field) and using different naming conventions. VirtualDB uses an external YAML configuration to map these varying structures to a common schema, normalize factor level names (e.g., “D-glucose”, “dextrose”, “glu” all become “glucose”), and enable cross-dataset queries with standardized field names and values.

For primary datasets, VirtualDB creates:

  • <db_name>_meta – one row per sample with derived metadata columns
  • <db_name> – full measurement-level data joined to the metadata view

For comparative analysis datasets, VirtualDB creates:

  • <db_name>_expanded – the raw data with composite ID fields parsed into <link_field>_source (aliased to configured db_name) and <link_field>_id (sample_id) columns

See the configuration guide for setup details and the tutorial for usage examples.

Advanced Usage

The underlying DuckDB connection is available as vdb._conn. You can use _conn to execute any SQL on the database, eg creating more views, or creating a table in memory.

Custom views created this way appear in tables(), describe(), and get_fields() automatically because those methods query DuckDB’s information_schema. Custom tables do not appear in tables() (which only lists views), but are fully queryable via vdb.query().

Example – create a materialized analysis table::

# Create a persistent in-memory table from a complex query.
# This example selects one "best" Hackett-2020 sample per regulator
# using a priority system: ZEV+P > GEV+P > GEV+M.
vdb._conn.execute("""
    CREATE OR REPLACE TABLE hackett_analysis_set AS
    WITH regulator_tiers AS (
        SELECT
            regulator_locus_tag,
            CASE
                WHEN BOOL_OR(mechanism = 'ZEV' AND restriction = 'P') THEN 1
                WHEN BOOL_OR(mechanism = 'GEV' AND restriction = 'P') THEN 2
                ELSE 3
            END AS tier
        FROM hackett_meta
        WHERE regulator_locus_tag NOT IN ('Z3EV', 'GEV')
        GROUP BY regulator_locus_tag
    ),
    tier_filter AS (
        SELECT
            h.sample_id, h.regulator_locus_tag, h.regulator_symbol,
            h.mechanism, h.restriction, h.date, h.strain, t.tier
        FROM hackett_meta h
        JOIN regulator_tiers t USING (regulator_locus_tag)
        WHERE
            (t.tier = 1 AND h.mechanism = 'ZEV' AND h.restriction = 'P')
            OR (t.tier = 2 AND h.mechanism = 'GEV' AND h.restriction = 'P')
            OR (t.tier = 3 AND h.mechanism = 'GEV' AND h.restriction = 'M')
    )
    SELECT DISTINCT
        sample_id, regulator_locus_tag, regulator_symbol,
        mechanism, restriction, date, strain
    FROM tier_filter
    WHERE regulator_symbol NOT IN ('GCN4', 'RDS2', 'SWI1', 'MAC1')
    ORDER BY regulator_locus_tag, sample_id
""")

df = vdb.query("SELECT * FROM hackett_analysis_set")

Tables and views created this way are in-memory only and do not persist across VirtualDB instances. They exist for the lifetime of the DuckDB connection.

API Reference

labretriever.virtual_db.VirtualDB

A query interface across heterogeneous datasets.

DuckDB views are lazily registered over Parquet files on first query() call. The user writes SQL against named views.

Attributes:

Name Type Description
config

Validated MetadataConfig

token

Optional HuggingFace token

Source code in labretriever/virtual_db.py
 189
 190
 191
 192
 193
 194
 195
 196
 197
 198
 199
 200
 201
 202
 203
 204
 205
 206
 207
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
class VirtualDB:
    """
    A query interface across heterogeneous datasets.

    DuckDB views are lazily registered over Parquet files on first
    ``query()`` call. The user writes SQL against named views.

    :ivar config: Validated MetadataConfig
    :ivar token: Optional HuggingFace token

    """

    def __init__(
        self,
        config_path: Path | str,
        token: str | None = None,
        duckdb_connection: duckdb.DuckDBPyConnection | None = None,
        local_files_only: bool = False,
        cache_dir: Path | str | None = None,
    ):
        """
        Initialize VirtualDB with configuration.

        Creates the DuckDB connection and registers all views immediately.

        :param config_path: Path to YAML configuration file
        :param token: Optional HuggingFace token for private datasets
        :param duckdb_connection: Optional DuckDB connection. If provided, views will be
            registered on this connection instead of creating a new in-memory database.
            This provides a method of using a persistent database file. If not provided,
            an in-memory DuckDB connection is created.
        :param local_files_only: If ``True``, skip HuggingFace network checks and use
            only locally cached files. Eliminates per-dataset ``repo_info``
            HTTP round-trips on warm restarts. Raises
            ``huggingface_hub.utils.LocalEntryNotFoundError`` if
            any required file is absent from the local cache.
        :param cache_dir: Override the HuggingFace cache directory. When ``None``,
            ``huggingface_hub`` resolves the location from ``HF_CACHE_DIR`` /
            ``HF_HOME`` / its own default. Pass an explicit path to store or read
            parquet snapshots from a non-default location (e.g. a bundled directory
            inside the deployed application).
        :raises FileNotFoundError: If config file does not exist
        :raises ValueError: If configuration is invalid

        """
        self.config = MetadataConfig.from_yaml(config_path)
        self.token = token
        self.local_files_only = local_files_only
        # Explicit argument wins; fall back to get_cache_dir() which reads
        # HF_CACHE_DIR at call time so CLI --cache-dir is respected even when
        # VirtualDB is constructed without the parameter.
        from labretriever.constants import get_cache_dir

        self.cache_dir: Path = (
            Path(cache_dir) if cache_dir is not None else get_cache_dir()
        )

        self._conn: duckdb.DuckDBPyConnection = (
            duckdb_connection
            if duckdb_connection is not None
            else duckdb.connect(":memory:")
        )

        # db_name -> (repo_id, config_name)
        self.db_name_map = self._build_db_name_map()

        # Prepared queries: name -> sql
        self._prepared_queries: dict[str, str] = {}

        _t0 = time.monotonic()

        t = time.monotonic()
        self._load_datacards()
        logger.debug("_load_datacards completed in %.3fs", time.monotonic() - t)

        t = time.monotonic()
        self._validate_datacards()
        logger.debug("_validate_datacards completed in %.3fs", time.monotonic() - t)

        t = time.monotonic()
        self._update_cache()
        logger.debug("_update_cache completed in %.3fs", time.monotonic() - t)

        t = time.monotonic()
        self._register_all_views()
        logger.debug("_register_all_views completed in %.3fs", time.monotonic() - t)

        t = time.monotonic()
        self._build_column_metadata()
        logger.debug("_build_column_metadata completed in %.3fs", time.monotonic() - t)

        logger.debug("VirtualDB.__init__ total: %.3fs", time.monotonic() - _t0)

    # ------------------------------------------------------------------
    # Public API
    # ------------------------------------------------------------------

    def query(self, sql: str, **params: Any) -> pd.DataFrame:
        """
        Execute SQL or a prepared query and return a DataFrame.

        If *sql* matches a registered prepared-query name the stored
        SQL template is used instead. Keyword arguments are passed as
        named parameters to DuckDB.

        :param sql: Raw SQL string **or** name of a prepared query
        :param params: Named parameters (DuckDB ``$name`` syntax)
        :return: Query result as a pandas DataFrame

        Examples::

            # Raw SQL
            df = vdb.query("SELECT * FROM harbison LIMIT 5")

            # With parameters
            df = vdb.query(
                "SELECT * FROM harbison_meta WHERE carbon_source = $cs",
                cs="glucose",
            )

            # Prepared query
            vdb.prepare("top", "SELECT * FROM harbison_meta LIMIT $n")
            df = vdb.query("top", n=10)

        """
        # param `sql` may be a prepared query name, a raw sql statement, or
        # a parameterized sql statement that is not prepared. If it exists as a key
        # in the _prepared_queries dict, we use the prepared sql. Otherwise, we
        # use the sql as passed to query().
        resolved = self._prepared_queries.get(sql, sql)
        try:
            if params:
                return self._conn.execute(resolved, params).fetchdf()
            return self._conn.execute(resolved).fetchdf()
        except Exception as exc:
            import pprint

            params_repr = pprint.pformat(params, indent=2)
            raise QueryError(
                f"query failed: {exc}\n\n" f"SQL:\n{sql}\n\n" f"params:\n{params_repr}"
            ) from exc

    def prepare(self, name: str, sql: str, overwrite: bool = False) -> None:
        """
        Register a named parameterized query for later use.

        Parameters use DuckDB ``$name`` syntax.

        :param name: Query name (must not collide with a view name)
        :param sql: SQL template with ``$name`` parameters
        :param overwrite: If True, overwrite existing prepared query
            with same name
        :raises ValueError: If *name* collides with an existing view

        Example::

            vdb.prepare("glucose_regs", '''
                SELECT regulator_symbol, COUNT(*) AS n
                FROM harbison_meta
                WHERE carbon_source = $cs
                GROUP BY regulator_symbol
                HAVING n >= $min_n
            ''')
            df = vdb.query("glucose_regs", cs="glucose", min_n=2)

        """

        if name in self._list_views() and not overwrite:
            error_msg = (
                f"Prepared-query name '{name}' collides with "
                f"an existing view. Choose a different name or set "
                f"overwrite=True."
            )
            logger.error(error_msg)
            raise ValueError(error_msg)
        self._prepared_queries[name] = sql

    def tables(self) -> list[str]:
        """
        Return sorted list of registered view names.

        :return: Sorted list of view names

        """

        return sorted(self._list_views())

    def describe(self, table: str | None = None) -> pd.DataFrame:
        """
        Describe column names and types for one or all views.

        :param table: View name, or None for all views
        :return: DataFrame with columns ``table``, ``column_name``,
                 ``column_type``

        """

        if table is not None:
            df = self._conn.execute(f"DESCRIBE {table}").fetchdf()
            df.insert(0, "table", table)
            return df

        frames = []
        for view in sorted(self._list_views()):
            df = self._conn.execute(f"DESCRIBE {view}").fetchdf()
            df.insert(0, "table", view)
            frames.append(df)
        if not frames:
            return pd.DataFrame(columns=["table", "column_name", "column_type"])
        return pd.concat(frames, ignore_index=True)

    def get_fields(self, table: str | None = None) -> list[str]:
        """
        Return column names for a view or all unique columns.

        :param table: View name, or None for all views
        :return: Sorted list of column names

        """

        if table is not None:
            cols = self._conn.execute(
                f"SELECT column_name FROM information_schema.columns "
                f"WHERE table_name = '{table}'"
            ).fetchdf()
            return sorted(cols["column_name"].tolist())

        all_cols: set[str] = set()
        for view in self._list_views():
            cols = self._conn.execute(
                f"SELECT column_name FROM information_schema.columns "
                f"WHERE table_name = '{view}'"
            ).fetchdf()
            all_cols.update(cols["column_name"].tolist())
        return sorted(all_cols)

    def get_common_fields(self) -> list[str]:
        """
        Return columns present in ALL primary ``_meta`` views.

        Primary dataset views are those without ``links`` in their
        config (i.e. not comparative datasets).

        :return: Sorted list of common column names

        """

        meta_views = self._get_primary_meta_view_names()
        if not meta_views:
            return []

        sets = []
        for view in meta_views:
            cols = self._conn.execute(
                f"SELECT column_name FROM information_schema.columns "
                f"WHERE table_name = '{view}'"
            ).fetchdf()
            sets.append(set(cols["column_name"].tolist()))

        common = set.intersection(*sets)
        return sorted(common)

    def get_datasets(self) -> list[str]:
        """
        Return the sorted list of dataset names known to this VirtualDB.

        Dataset names are the resolved ``db_name`` values from the
        configuration (falling back to the config_name when ``db_name``
        is not explicitly set). These are the names accepted by
        :meth:`get_tags` and queryable via :meth:`query`.

        Unlike :meth:`tables`, this method reads directly from the
        configuration and does not require views to be registered, so
        no data is downloaded.

        :return: Sorted list of dataset names

        """
        return sorted(self.db_name_map)

    def materialize(self) -> None:
        """
        Replace all registered dataset views with in-memory DuckDB tables.

        For each dataset known to this VirtualDB, reads the public data view
        and the corresponding ``_meta`` view into in-memory tables, then
        re-points the original view names at those tables. Subsequent queries
        hit RAM instead of re-scanning parquet files, which substantially
        reduces per-query latency at the cost of increased startup time and
        memory usage.

        Only views that exist in the current DuckDB session are materialized;
        missing views are silently skipped. Safe to call multiple times —
        uses ``CREATE OR REPLACE`` semantics.

        """
        conn = self._conn
        for db_name in self.get_datasets():
            for view_name in (db_name, f"{db_name}_meta"):
                row = conn.execute(
                    "SELECT view_name FROM duckdb_views() WHERE view_name = ?",
                    [view_name],
                ).fetchone()
                if row is None:
                    continue
                table_name = f"_mat_{view_name}"
                conn.execute(
                    f"CREATE OR REPLACE TABLE {table_name} AS SELECT * FROM {view_name}"
                )
                conn.execute(
                    f"CREATE OR REPLACE VIEW {view_name} AS SELECT * FROM {table_name}"
                )
                logger.debug("materialize: %s -> %s", view_name, table_name)

    def get_tags(self, db_name: str) -> dict[str, str]:
        """
        Return the merged tags for a dataset.

        Tags are defined in the configuration at the repository and/or
        dataset level. Dataset-level tags override repository-level tags
        with the same key. See the ``tags`` section of the configuration
        guide for details.

        :param db_name: Dataset name as it appears in :meth:`tables` (the
            resolved ``db_name`` from the configuration, or the
            ``config_name`` if ``db_name`` was not explicitly set).
        :return: Dict of merged tags, or empty dict if the dataset has no
            tags or the name is not found.

        """
        if db_name not in self.db_name_map:
            return {}
        repo_id, config_name = self.db_name_map[db_name]
        return self.config.get_tags(repo_id, config_name)

    def get_condition_field_info(self, db_name: str) -> dict[str, Any] | None:
        """
        Return hierarchically linked property column groups for a dataset.

        Some datasets expose multiple derived property columns in their
        ``_meta`` view (e.g. ``"Carbon source"`` and ``"Temperature"``) that
        are both extracted from the same per-sample experimental-condition
        field via field+path ``PropertyMapping`` entries in the VirtualDB
        configuration.  Because these columns are co-derived from the same
        source field, the values that can appear in one column are constrained
        by the value selected in another (selecting ``Carbon source =
        glucose`` limits which ``Temperature`` values are meaningful).

        This method identifies such groups and enriches them with per-level
        descriptions from the DataCard, making it straightforward for
        downstream tools to build conditional filter interfaces.

        :param db_name: Dataset name as returned by :meth:`get_datasets`.
        :type db_name: str
        :returns: A dict keyed by source parquet field name.  Each value is
            a dict with two keys:

            ``"property_cols"`` — ``list[str]``
                The derived property column names that appear in the
                ``{db_name}_meta`` view and share this source field.

            ``"level_descriptions"`` — ``dict[str, str]``
                Maps each categorical level of the source field to a
                human-readable description string retrieved from the
                DataCard via :meth:`~labretriever.datacard.DataCard.get_field_definitions`. # noqa: E501
                Levels without a ``"description"`` key in their definition
                dict default to ``"Description unavailable"``.

            Returns ``None`` when the dataset has no linked property groups
            (no Type-A field-only mapping exists, or there are no other
            property columns to serve as downstream members),
            or when ``db_name`` is not found.
        :rtype: dict[str, Any] | None

        Example::

            info = vdb.get_condition_field_info("harbison")
            if info is not None:
                for src_field, group in info.items():
                    print(f"Source field '{src_field}' links:")
                    print(f"  property columns: {group['property_cols']}")
                    for level, desc in group["level_descriptions"].items():
                        print(f"  {level}: {desc}")

        """
        warnings.warn(
            "get_condition_field_info is deprecated; use get_column_metadata instead.",
            DeprecationWarning,
            stacklevel=2,
        )
        if db_name not in self.db_name_map:
            return None

        repo_id, config_name = self.db_name_map[db_name]
        mappings = self.config.get_property_mappings(repo_id, config_name)

        # Collect Type-C (path-only) columns — repo-level experimental condition
        # extractions.  These are shared downstream members for any Type-A anchor
        # that has at least one co-occurring Type-B sibling or Type-C companion.
        type_c_cols = [
            prop_col
            for prop_col, pm in mappings.items()
            if pm.field is None and pm.path is not None
        ]

        # Build per-src_field groups anchored by Type-A (field-only) mappings.
        # A Type-A mapping is a condition anchor only when its prop_col name
        # differs from pm.field — this distinguishes semantic aliases like
        # "Experimental condition" (field=condition) from identity renames like
        # "regulator_locus_tag" (field=regulator_locus_tag).
        # The group requires at least one downstream member: a Type-B col sharing
        # the same src_field, or a Type-C col (when no Type-B siblings exist).
        linked: dict[str, dict[str, list[str]]] = {}
        for prop_col, pm in mappings.items():
            if pm.field is None or pm.path is not None:
                continue  # not Type-A
            # Exclude identity renames: prop_col and pm.field are the same name,
            # possibly differing only in case or whitespace (e.g. "Regulator locus tag"
            # -> "regulator_locus_tag").  Normalise both sides before comparing.
            if prop_col.lower().replace(" ", "_") == pm.field.lower().replace(" ", "_"):
                continue
            src_field = pm.field
            type_b_for_field = [
                col
                for col, m in mappings.items()
                if m.field == src_field and m.path is not None
            ]
            type_c_for_group = [c for c in type_c_cols if c not in type_b_for_field]
            downstream = type_b_for_field + type_c_for_group
            if downstream:
                linked[src_field] = {
                    "prop_cols": downstream,
                    "type_b_cols": type_b_for_field,
                }

        card = self.datacards.get(repo_id)

        # Fallback: when no Type-A aliases exist but there are Type-C downstream
        # cols, use the DataCard's condition_fields (role=experimental_condition)
        # as implicit anchors.  Use the first condition field only; multiple
        # condition fields sharing identical downstream cols would produce
        # duplicate input IDs in the UI.
        # Fallback: DataCard condition_fields as implicit anchors.  All Type-C
        # cols are downstream; none are Type-B (no per-level field derivation).
        if not linked and type_c_cols and card is not None:
            try:
                schema = card.extract_metadata_schema(config_name)
                cond_fields = schema.get("condition_fields", [])
                if cond_fields:
                    linked[cond_fields[0]] = {
                        "prop_cols": list(type_c_cols),
                        "type_b_cols": [],
                    }
            except Exception:
                pass

        if not linked:
            return None

        # Resolve the config name to try for field definitions.  For datasets
        # with external metadata, the definitions live in the _meta config rather
        # than the primary annotated-features config, so prefer that when present.
        meta_config_name = self._external_meta_configs.get(db_name, config_name)

        result: dict[str, Any] = {}
        for src_field, col_info in linked.items():
            prop_cols = col_info["prop_cols"]
            type_b_cols = col_info["type_b_cols"]
            level_descriptions: dict[str, str] = {}
            if card is not None:
                for _cfg in (meta_config_name, config_name):
                    try:
                        defs = card.get_field_definitions(_cfg, src_field)
                        if defs:
                            for level_val, definition in defs.items():
                                desc = (
                                    definition.get(
                                        "description", "Description unavailable"
                                    )
                                    if isinstance(definition, dict)
                                    else "Description unavailable"
                                )
                                level_descriptions[str(level_val)] = desc
                            break  # found definitions — stop trying configs
                    except Exception:
                        logger.debug(
                            "No field definitions for %s/%s field '%s'",
                            db_name,
                            _cfg,
                            src_field,
                        )
            result[src_field] = {
                "property_cols": prop_cols,
                "type_b_cols": type_b_cols,
                "level_descriptions": level_descriptions,
            }
        return result

    def get_column_metadata(self, db_name: str) -> dict[str, ColumnMeta] | None:
        """
        Return per-column metadata for a primary dataset.

        Metadata is collected from the DataCard during VirtualDB construction
        and includes the feature description, semantic role, and per-level
        definitions for ``experimental_condition`` columns.

        :param db_name: Dataset name as registered in the VirtualDB config.
        :returns: Dict mapping column name to :class:`ColumnMeta`, or ``None``
            if ``db_name`` is not found or has no recorded metadata.
        :rtype: dict[str, ColumnMeta] | None

        """
        return self._column_metadata.get(db_name)

    def get_dataset_description(self, db_name: str) -> str | None:
        """
        Return the description for a dataset.

        The VirtualDB config ``description`` field takes precedence over the
        DataCard config description. Returns ``None`` if neither is defined.

        :param db_name: Dataset name as registered in the VirtualDB config.
        :returns: Description string, or ``None`` if not found.
        :rtype: str | None

        """
        if db_name not in self.db_name_map:
            return None
        repo_id, config_name = self.db_name_map[db_name]
        # Check VirtualDB config description override first
        repo_cfg = self.config.repositories.get(repo_id)
        if repo_cfg and repo_cfg.dataset:
            vdb_dataset_cfg = repo_cfg.dataset.get(config_name)
            if vdb_dataset_cfg and vdb_dataset_cfg.description is not None:
                return vdb_dataset_cfg.description
        # Fall back to DataCard description
        card = self.datacards.get(repo_id)
        if card is None:
            return None
        cfg = card.get_config(config_name)
        return cfg.description if cfg is not None else None

    def get_citation(self, db_name: str) -> str | None:
        """
        Return the citation for a dataset.

        The dataset-level citation takes precedence over the repository-level
        citation. If neither is defined, returns ``None``.

        :param db_name: Dataset name as registered in the VirtualDB config.
        :returns: Citation string, or ``None`` if no citation is defined.
        :rtype: str | None

        """
        if db_name not in self.db_name_map:
            return None
        repo_id, config_name = self.db_name_map[db_name]
        card = self.datacards.get(repo_id)
        if card is None:
            return None
        return card.get_citation(config_name)

    # ------------------------------------------------------------------
    # Initialisation phases
    # ------------------------------------------------------------------

    def _load_datacards(self) -> None:
        """
        Fetch (or load from cache) the DataCard for every distinct repo.

        Populates ``self.datacards`` keyed by ``repo_id``. Failures are
        logged as warnings and the repo is omitted from the dict so that
        subsequent phases can skip it gracefully.

        """
        self.datacards: dict[str, DataCard] = {}
        seen_repos: set[str] = set()
        for repo_id, _ in self.db_name_map.values():
            if repo_id in seen_repos:
                continue
            seen_repos.add(repo_id)
            try:
                _t = time.monotonic()
                self.datacards[repo_id] = _cached_datacard(repo_id, token=self.token)
                logger.debug(
                    "_load_datacards: %s loaded in %.3fs",
                    repo_id,
                    time.monotonic() - _t,
                )
            except Exception as exc:
                logger.warning(
                    "Could not load datacard for repo '%s': %s",
                    repo_id,
                    exc,
                )

    def _validate_datacards(self) -> None:
        """
        Cross-check the VirtualDB config against the loaded datacards.

        Checks that every dataset with a ``links`` field in the VirtualDB
        config has ``dataset_type: comparative`` in its HuggingFace datacard.
        Also resolves ``self._dataset_schemas`` and
        ``self._external_meta_configs`` (keyed by ``db_name``) for use by
        ``_update_cache`` and ``_register_all_views``.

        :raises ValueError: If a dataset with ``links`` does not have
            ``dataset_type: comparative`` in its datacard.

        """
        self._dataset_schemas: dict[str, DatasetSchema] = {}
        # db_name -> external metadata config_name (for applies_to datasets)
        self._external_meta_configs: dict[str, str] = {}

        for db_name, (repo_id, config_name) in self.db_name_map.items():
            repo_cfg = self.config.repositories.get(repo_id)
            ds_cfg = (
                repo_cfg.dataset.get(config_name)
                if repo_cfg and repo_cfg.dataset
                else None
            )
            card = self.datacards.get(repo_id)

            # Validate comparative dataset_type agreement.
            if ds_cfg and ds_cfg.links:
                if card is not None:
                    dc_config = card.get_config(config_name)
                    if (
                        dc_config is not None
                        and dc_config.dataset_type != DatasetType.COMPARATIVE
                    ):
                        raise ValueError(
                            f"Dataset '{config_name}' in repo '{repo_id}' has "
                            f"'links' in the VirtualDB config, indicating a "
                            f"comparative dataset, but the HuggingFace datacard "
                            f"declares dataset_type='{dc_config.dataset_type}'. "
                            f"Update the datacard to use dataset_type: comparative."
                        )
                continue  # comparative datasets need no schema resolution

            # Resolve dataset schema and external metadata config for
            # primary datasets.
            if card is None:
                continue
            try:
                schema = card.get_dataset_schema(config_name)
            except Exception as exc:
                logger.warning(
                    "Could not get dataset schema for %s/%s: %s",
                    repo_id,
                    config_name,
                    exc,
                )
                continue
            if schema is not None:
                self._dataset_schemas[db_name] = schema
            if (
                schema is not None
                and schema.metadata_source == "external"
                and schema.external_metadata_config
            ):
                self._external_meta_configs[db_name] = schema.external_metadata_config

    def _update_cache(self) -> None:
        """
        Download (or locate cached) Parquet files for all dataset configs.

        Populates ``self._parquet_files`` keyed by ``db_name``. For datasets
        with external metadata (identified during ``_validate_datacards``),
        also downloads those files and stores them under the key
        ``"__<db_name>_meta"`` so ``_register_all_views`` can read them
        without further network calls.

        """
        self._parquet_files: dict[str, list[str]] = {}
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            _t = time.monotonic()
            files = self._resolve_parquet_files(repo_id, config_name)
            self._parquet_files[db_name] = files
            logger.debug(
                "_update_cache: %s resolved %d file(s) in %.3fs",
                db_name,
                len(files),
                time.monotonic() - _t,
            )

        for db_name, ext_config_name in self._external_meta_configs.items():
            repo_id, _ = self.db_name_map[db_name]
            _t = time.monotonic()
            files = self._resolve_parquet_files(repo_id, ext_config_name)
            self._parquet_files[f"__{db_name}_meta"] = files
            logger.debug(
                "_update_cache: %s (ext meta) resolved %d file(s) in %.3fs",
                db_name,
                len(files),
                time.monotonic() - _t,
            )

    def _register_all_views(self) -> None:
        """
        Register all DuckDB views in dependency order.

        Expects ``self._parquet_files``, ``self._dataset_schemas``, and
        ``self._external_meta_configs`` to have been populated by the earlier
        init phases. No network or disk access occurs here.

        """
        # 1. Raw per-dataset views (internal __<db_name>_parquet
        # plus public <db_name> for primary datasets only)
        _t = time.monotonic()
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            comparative = self._is_comparative(repo_id, config_name)
            self._register_raw_view(
                db_name,
                parquet_only=comparative,
            )
        logger.debug(
            "_register_all_views: raw views completed in %.3fs", time.monotonic() - _t
        )

        # 2. External metadata parquet views.
        # When a data config's metadata lives in a separate HF config
        # (applies_to), register its parquet as __<db_name>_metadata_parquet.
        _t = time.monotonic()
        self._external_meta_views: dict[str, str] = {}
        for db_name, ext_config_name in self._external_meta_configs.items():
            meta_view = f"__{db_name}_metadata_parquet"
            files = self._parquet_files.get(f"__{db_name}_meta", [])
            if not files:
                logger.warning(
                    "No parquet files for external metadata config "
                    "'%s' (db_name '%s') -- skipping external metadata view",
                    ext_config_name,
                    db_name,
                )
                continue
            files_sql = ", ".join(f"'{f}'" for f in files)
            try:
                self._conn.execute(
                    f"CREATE OR REPLACE VIEW {meta_view} AS "
                    f"SELECT * FROM read_parquet([{files_sql}])"
                )
            except Exception as exc:
                logger.warning(
                    "Failed to create external metadata view '%s': %s",
                    meta_view,
                    exc,
                )
                continue
            self._external_meta_views[db_name] = meta_view
        logger.debug(
            "_register_all_views: external meta views completed in %.3fs",
            time.monotonic() - _t,
        )

        # 3. Metadata views for primary datasets (<db_name>_meta)
        _t = time.monotonic()
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            if not self._is_comparative(repo_id, config_name):
                self._register_meta_view(db_name, repo_id, config_name)
        logger.debug(
            "_register_all_views: meta views completed in %.3fs", time.monotonic() - _t
        )

        # 4. Replace primary raw views with join to _meta so
        # derived columns (e.g. carbon_source) are available
        _t = time.monotonic()
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            if not self._is_comparative(repo_id, config_name):
                self._enrich_raw_view(db_name)
        logger.debug(
            "_register_all_views: enrich raw views completed in %.3fs",
            time.monotonic() - _t,
        )

        # 5. Comparative expanded views (pre-parsed composite IDs)
        _t = time.monotonic()
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            repo_cfg = self.config.repositories.get(repo_id)
            if not repo_cfg or not repo_cfg.dataset:
                continue
            ds_cfg = repo_cfg.dataset.get(config_name)
            if ds_cfg and ds_cfg.links:
                self._register_comparative_expanded_view(db_name, ds_cfg)
        logger.debug(
            "_register_all_views: comparative views completed in %.3fs",
            time.monotonic() - _t,
        )

    def _build_column_metadata(self) -> None:
        """
        Collect per-column metadata from DataCards for all primary datasets.

        Populates ``self._column_metadata`` keyed by ``db_name``.  For each
        primary (non-comparative) dataset, features are gathered from the
        primary DataCard config and, when present, from the external metadata
        config (``_external_meta_configs``).  Property mappings are then
        walked to propagate metadata to output column names that differ from
        the underlying DataCard feature names (e.g. Type-A renames).

        """
        self._column_metadata: dict[str, dict[str, ColumnMeta]] = {}

        for db_name, (repo_id, config_name) in self.db_name_map.items():
            if self._is_comparative(repo_id, config_name):
                continue

            card = self.datacards.get(repo_id)
            if card is None:
                continue

            # Collect features from primary config and external meta config.
            feature_meta: dict[str, ColumnMeta] = {}
            configs_to_check: list[str] = [config_name]
            ext_cfg_name = self._external_meta_configs.get(db_name)
            if ext_cfg_name:
                configs_to_check.append(ext_cfg_name)

            for cfg in configs_to_check:
                try:
                    features = card.get_features(cfg)
                except Exception:
                    continue
                for feat in features:
                    level_defs: dict[str, str] | None = None
                    if feat.role == "experimental_condition" and feat.definitions:
                        level_defs = {
                            str(lv): (
                                d.get("description", "")
                                if isinstance(d, dict)
                                else str(d)
                            )
                            for lv, d in feat.definitions.items()
                        }
                    feature_meta[feat.name] = ColumnMeta(
                        description=feat.description,
                        role=feat.role,
                        level_definitions=level_defs,
                    )

            # Propagate metadata to Type-A rename output column names.
            # Type-A: field-only mapping (no path, no expression) — the output col
            # is a rename or alias of the source feature and inherits its metadata.
            # Type-B (field+path) and Type-C (path-only) produce derived columns
            # with independent semantics and do not inherit the source's role.
            try:
                mappings = self.config.get_property_mappings(repo_id, config_name)
            except Exception:
                mappings = {}

            for out_col, pm in mappings.items():
                if out_col in feature_meta:
                    continue  # already recorded under this name
                # Only propagate for Type-A (field-only, no path, no expression).
                if pm.field is None or pm.path is not None or pm.expression is not None:
                    continue
                src = pm.field
                if src in feature_meta:
                    feature_meta[out_col] = feature_meta[src]

            # Register stub ColumnMeta for Type-B and Type-C derived output columns.
            # These have no FeatureInfo of their own but exist in the _meta view.
            # A stub entry (role=None, level_definitions=None) ensures callers can
            # discover all queryable columns, e.g. to build cascade filter effects.
            for out_col in mappings:
                if out_col not in feature_meta:
                    feature_meta[out_col] = ColumnMeta()

            if feature_meta:
                self._column_metadata[db_name] = feature_meta

    # ------------------------------------------------------------------
    # db_name mapping
    # ------------------------------------------------------------------

    def _build_db_name_map(self) -> dict[str, tuple[str, str]]:
        """
        Build mapping from resolved db_name to (repo_id, config_name).

        :return: Dict mapping db_name -> (repo_id, config_name)

        """
        mapping: dict[str, tuple[str, str]] = {}
        for repo_id, repo_cfg in self.config.repositories.items():
            if not repo_cfg.dataset:
                continue
            for config_name, ds_cfg in repo_cfg.dataset.items():
                resolved = ds_cfg.db_name or config_name
                mapping[resolved] = (repo_id, config_name)
        return mapping

    # ------------------------------------------------------------------
    # Parquet file resolution
    # ------------------------------------------------------------------

    def _resolve_parquet_files(self, repo_id: str, config_name: str) -> list[str]:
        """
        Download (or locate cached) Parquet files for a dataset config.

        Uses ``huggingface_hub.snapshot_download`` with the file patterns
        from the DataCard.

        :param repo_id: HuggingFace repository ID
        :param config_name: Dataset configuration name
        :return: List of absolute paths to Parquet files

        """
        card = DataCard(repo_id, token=self.token)
        config = card.get_config(config_name)
        if not config:
            logger.warning(
                "Config '%s' not found in repo '%s'",
                config_name,
                repo_id,
            )
            return []

        file_patterns = [df.path for df in config.data_files]

        from huggingface_hub import snapshot_download

        logger.debug(
            "snapshot_download start: repo=%s patterns=%s", repo_id, file_patterns
        )
        t0 = time.monotonic()
        downloaded_path = snapshot_download(
            repo_id=repo_id,
            repo_type="dataset",
            allow_patterns=file_patterns,
            token=self.token,
            local_files_only=self.local_files_only,
            cache_dir=self.cache_dir,
        )
        elapsed = time.monotonic() - t0
        logger.debug(
            "snapshot_download done: repo=%s elapsed=%.3fs path=%s",
            repo_id,
            elapsed,
            downloaded_path,
        )

        parquet_files: list[str] = []
        for pattern in file_patterns:
            file_path = Path(downloaded_path) / pattern
            if file_path.exists() and file_path.suffix == ".parquet":
                parquet_files.append(str(file_path))
            elif "*" in pattern:
                base = Path(downloaded_path)
                parquet_files.extend(
                    str(f) for f in base.glob(pattern) if f.suffix == ".parquet"
                )
            else:
                parent_dir = Path(downloaded_path) / Path(pattern).parent
                if parent_dir.exists():
                    parquet_files.extend(str(f) for f in parent_dir.glob("*.parquet"))

        return parquet_files

    # ------------------------------------------------------------------
    # View registration helpers
    # ------------------------------------------------------------------

    def _register_raw_view(
        self,
        db_name: str,
        *,
        parquet_only: bool = False,
    ) -> None:
        """
        Register a raw DuckDB view over pre-resolved Parquet files.

        Creates an internal ``__<db_name>_parquet`` view that reads
        directly from the Parquet files. For primary datasets, also
        creates a public ``<db_name>`` view (initially identical)
        that may later be replaced by ``_enrich_raw_view``.

        For comparative datasets, only the internal parquet view is
        created; the public view is the ``_expanded`` view instead.

        Parquet files must have been resolved by ``_update_cache``
        before this method is called.

        :param db_name: View name
        :param parquet_only: If True, only create the internal
            ``__<db_name>_parquet`` view (no public ``<db_name>``).

        """
        files = self._parquet_files.get(db_name, [])
        if not files:
            logger.warning(
                "No parquet files for db_name '%s' -- skipping view",
                db_name,
            )
            return

        files_sql = ", ".join(f"'{f}'" for f in files)
        parquet_sql = f"SELECT * FROM read_parquet([{files_sql}])"
        self._conn.execute(
            f"CREATE OR REPLACE VIEW __{db_name}_parquet AS " f"{parquet_sql}"
        )
        if not parquet_only:
            sample_col = self._get_sample_id_col(db_name)
            if sample_col == "sample_id":
                public_select = f"SELECT * FROM __{db_name}_parquet"
            else:
                raw_cols = self._get_view_columns(f"__{db_name}_parquet")
                parts: list[str] = []
                for col in raw_cols:
                    if col == sample_col:
                        parts.append(f"{col} AS sample_id")
                    elif col == "sample_id":
                        parts.append(f"{col} AS sample_id_orig")
                    else:
                        parts.append(col)
                cols_sql = ", ".join(parts)
                public_select = f"SELECT {cols_sql} FROM __{db_name}_parquet"
            self._conn.execute(f"CREATE OR REPLACE VIEW {db_name} AS {public_select}")

    def _register_meta_view(self, db_name: str, repo_id: str, config_name: str) -> None:
        """
        Register a ``<db_name>_meta`` view with one row per sample.

        Includes metadata columns from the DataCard plus any derived columns
        from config property mappings (resolved against DataCard definitions
        with factor aliases applied).

        For datasets with external metadata (a separate HF config with
        ``applies_to``), JOINs the data parquet to the metadata parquet
        on the configured sample_id column. The actual columns in the metadata
        parquet are determined by DuckDB introspection (``DESCRIBE``) rather
        than the DataCard feature list, because DataCard feature lists are
        conceptual schemas that may include columns not physically present
        in the parquet files.

        :param db_name: Base view name for the primary dataset
        :param repo_id: Repository ID
        :param config_name: Configuration name

        raises ValueError: If no metadata fields are found.
        raises BinderException: If view creation fails, with SQL details.

        """
        _t0 = time.monotonic()
        parquet_view = f"__{db_name}_parquet"
        if not self._view_exists(parquet_view):
            return

        sample_col = self._get_sample_id_col(db_name)

        # Pull ext_meta_view early -- needed for both meta_cols and
        # FROM clause construction.
        schema: DatasetSchema | None = self._dataset_schemas.get(db_name)
        ext_meta_view: str | None = self._external_meta_views.get(db_name)

        is_external = (
            ext_meta_view is not None
            and schema is not None
            and schema.metadata_source == "external"
        )

        if is_external:
            # DataCard feature lists are conceptual -- columns listed there
            # may not be physically present in the parquet file. Use DuckDB
            # introspection to get the actual columns in the metadata parquet.
            assert ext_meta_view is not None
            actual_meta_cols: set[str] = set(self._get_view_columns(ext_meta_view))
            meta_cols: list[str] = sorted(actual_meta_cols)
        elif schema is not None:
            actual_meta_cols = schema.metadata_columns
            meta_cols = sorted(actual_meta_cols)
        else:
            meta_cols = self._resolve_metadata_fields(repo_id, config_name) or []
            actual_meta_cols = set(meta_cols)

        if not meta_cols:
            raise ValueError(
                f"No metadata fields found for {repo_id}/{config_name}. "
                f"Cannot create meta view '{db_name}_meta'."
            )

        # FROM clause: JOIN data + metadata parquets when external,
        # plain parquet view otherwise.
        if is_external:
            assert ext_meta_view is not None
            # Use the configured sample_id column as the join key.
            # The DataCard feature intersection (schema.join_columns)
            # is unreliable because a data config's feature list may
            # document columns that are physically only in the metadata
            # parquet (present conceptually after a join, not in the
            # physical data parquet file).
            from_clause = (
                f"{parquet_view} d " f"JOIN {ext_meta_view} m " f"USING ({sample_col})"
            )
            is_join = True
        else:
            from_clause = parquet_view
            is_join = False

        def qualify(col: str) -> str:
            """Return qualified column name for JOIN context."""
            if not is_join:
                return col
            if col == sample_col:
                return col  # USING makes join key unqualified
            # Use the actual metadata parquet columns (from DuckDB
            # introspection) to decide qualification, not the DataCard
            # feature list which may be inaccurate.
            if col in actual_meta_cols:
                return f"m.{col}"
            return f"d.{col}"

        # Resolve derived property expressions first.
        # When a factor mapping has the same output name as its source
        # field (e.g. time -> time), the raw column must be renamed to
        # avoid a duplicate column name in the SELECT.  The rename uses
        # "<col>_orig", or "<col>_orig_1", etc., to avoid collisions with
        # other columns that already exist in the parquet.
        prop_result = self._resolve_property_columns(repo_id, config_name)

        # Collect all column names that exist in the parquet so we can
        # find a unique _orig suffix when needed.
        all_parquet_cols: set[str] = set(self._get_view_columns(parquet_view))

        # Map: raw_col -> alias_in_select for factor-overridden cols
        factor_col_renames: dict[str, str] = {}
        if prop_result is not None:
            _derived_exprs, _prop_raw_cols = prop_result
            for expr in _derived_exprs:
                # Detect factor CAST expressions of the form:
                # CAST(CAST(<field> AS VARCHAR) AS _enum_<key>) AS <key>
                # where <field> == <key> (in-place factor override).
                # The output column name is the last " AS <name>" token.
                parts = expr.rsplit(" AS ", 1)
                if len(parts) != 2:
                    continue
                # Strip double-quotes added by _quote_ident so we can
                # compare the bare name against parquet column names.
                out_col = parts[1].strip().strip('"')
                # Extract the innermost source field from the CAST chain.
                # Handles both:
                #   CAST(CAST(<field> AS VARCHAR) AS _enum_<key>)
                #   CAST(CAST(CAST(<field> AS BIGINT) AS VARCHAR) AS _enum_<key>)
                m = re.match(
                    r"CAST\(CAST\((?:CAST\()?(\w+)(?:\s+AS\s+BIGINT\))?"
                    r"\s+AS\s+VARCHAR\)\s+AS\s+_enum_\w+\)",
                    parts[0],
                )
                if m is None:
                    continue
                src_field = m.group(1)
                if src_field == out_col and out_col in all_parquet_cols:
                    # Find a unique _orig name
                    candidate = f"{out_col}_orig"
                    n = 1
                    while candidate in all_parquet_cols or candidate in (
                        v for v in factor_col_renames.values()
                    ):
                        candidate = f"{out_col}_orig_{n}"
                        n += 1
                    factor_col_renames[src_field] = candidate

        # Build SELECT: sample_id + metadata cols (deduplicated).
        # Raw columns that are factor-overridden are emitted with their
        # _orig alias instead of their original name.
        # If the configured sample_id column differs from "sample_id",
        # rename it so all views expose a consistent "sample_id" column.
        # If the parquet also has a literal "sample_id" column, preserve
        # it as "sample_id_orig" to avoid losing data.
        seen: set[str] = set()
        select_parts: list[str] = []
        rename_sample = sample_col != "sample_id"

        def add_col(col: str) -> None:
            if col in seen:
                return
            seen.add(col)
            alias = factor_col_renames.get(col)
            if alias:
                select_parts.append(f"{qualify(col)} AS {_quote_ident(alias)}")
            elif rename_sample and col == sample_col:
                select_parts.append(f"{qualify(col)} AS sample_id")
            elif rename_sample and col == "sample_id":
                select_parts.append(f"{qualify(col)} AS sample_id_orig")
            else:
                select_parts.append(qualify(col))

        add_col(sample_col)
        # When renaming, check if the parquet source also has a literal
        # "sample_id" column; if so, preserve it as "sample_id_orig".
        if rename_sample:
            source_cols = set(self._get_view_columns(parquet_view))
            if "sample_id" in source_cols:
                add_col("sample_id")
        for col in meta_cols:
            add_col(col)

        # Add derived property expressions from the VirtualDB config
        if prop_result is not None:
            derived_exprs, prop_raw_cols = prop_result
            # Ensure source columns needed by expressions are selected.
            # For external metadata datasets, restrict to columns physically
            # present in the metadata parquet -- data columns must not bleed
            # into the meta view.
            allowed_raw_cols = (
                [c for c in prop_raw_cols if c in actual_meta_cols]
                if is_external
                else prop_raw_cols
            )
            for col in allowed_raw_cols:
                add_col(col)
            # Rewrite CAST expressions to use the _orig alias when the
            # source field was renamed to avoid collision.
            if factor_col_renames:
                rewritten = []
                for expr in derived_exprs:
                    for orig, alias in factor_col_renames.items():
                        # Replace "CAST(<orig> AS" with "CAST(<alias> AS"
                        expr = expr.replace(f"CAST({orig} AS", f"CAST({alias} AS")
                    rewritten.append(expr)
                derived_exprs = rewritten
            # Qualify source column references inside expressions.
            # Covers:
            #   - simple alias:   "field AS ..."  → "m.field AS ..."
            #   - CAST alias:     "CAST(field AS ..." → "CAST(m.field AS ..."
            #   - CASE WHEN:      "CASE field WHEN..." / "field = ..."
            if is_join:
                qualified_exprs = []
                for expr in derived_exprs:
                    for raw_col in prop_raw_cols:
                        q = qualify(raw_col)
                        if q != raw_col:
                            expr = (
                                expr
                                # simple alias: bare field at start before " AS"
                                .replace(f"{raw_col} AS ", f"{q} AS ")
                                # CAST alias: field inside CAST(
                                .replace(f"CAST({raw_col} AS", f"CAST({q} AS")
                                # CASE WHEN patterns
                                .replace(f"CASE {raw_col} ", f"CASE {q} ").replace(
                                    f" {raw_col} = ", f" {q} = "
                                )
                            )
                    qualified_exprs.append(expr)
                derived_exprs = qualified_exprs
            select_parts.extend(derived_exprs)

        cols_sql = ", ".join(select_parts)
        sql = (
            f"CREATE OR REPLACE VIEW {db_name}_meta AS "
            f"SELECT DISTINCT {cols_sql} FROM {from_clause}"
        )
        try:
            self._conn.execute(sql)
        except BinderException as exc:
            raise BinderException(
                f"Failed to create meta view '{db_name}_meta'.\n"
                f"  schema: {schema}\n"
                f"  from_clause: {from_clause}\n"
                f"  SQL: {sql}\n"
                f"  error: {exc}"
            ) from exc
        logger.debug(
            "_register_meta_view: %s_meta completed in %.3fs",
            db_name,
            time.monotonic() - _t0,
        )

    def _enrich_raw_view(self, db_name: str) -> None:
        """
        Replace a primary raw view with a join to its ``_meta`` view.

        If ``<db_name>_meta`` has derived columns not present in the
        raw parquet view, recreates ``<db_name>`` as a join so derived
        columns (e.g. ``carbon_source``) appear alongside measurement
        data.

        :param db_name: Base view name for the primary dataset

        """
        meta_name = f"{db_name}_meta"
        parquet_name = f"__{db_name}_parquet"
        if not self._view_exists(meta_name) or not self._view_exists(parquet_name):
            return

        raw_cols_list = self._get_view_columns(parquet_name)
        raw_cols = set(raw_cols_list)
        meta_cols = set(self._get_view_columns(meta_name))

        sample_col = self._get_sample_id_col(db_name)
        rename_sample = sample_col != "sample_id"

        # Columns to pull from _meta that aren't already in raw parquet,
        # accounting for the sample_id rename: when renaming, "sample_id"
        # will appear in meta_cols (as the renamed column) but not in
        # raw_cols (which has the original name), so we must exclude it
        # from extra_cols since the rename in the raw SELECT already
        # provides it.
        if rename_sample:
            # "sample_id" and "sample_id_orig" come from the raw SELECT
            # rename, not from meta
            extra_cols = meta_cols - raw_cols - {"sample_id", "sample_id_orig"}
        else:
            extra_cols = meta_cols - raw_cols

        if not extra_cols:
            # No derived columns to add -- the view created in
            # _register_raw_view (which already handles the rename)
            # is sufficient.
            return

        if rename_sample:
            # Build explicit SELECT to rename the sample column
            raw_parts: list[str] = []
            for col in raw_cols_list:
                if col == sample_col:
                    raw_parts.append(f"r.{col} AS sample_id")
                elif col == "sample_id":
                    raw_parts.append(f"r.{col} AS sample_id_orig")
                else:
                    raw_parts.append(f"r.{col}")
            raw_select = ", ".join(raw_parts)
        else:
            raw_select = "r.*"

        if extra_cols:
            extra_select = ", ".join(f"m.{_quote_ident(c)}" for c in sorted(extra_cols))
            full_select = f"{raw_select}, {extra_select}"
        else:
            full_select = raw_select

        if rename_sample:
            join_clause = f"JOIN {meta_name} m ON r.{sample_col} = m.sample_id"
        else:
            join_clause = f"JOIN {meta_name} m USING ({sample_col})"

        self._conn.execute(
            f"CREATE OR REPLACE VIEW {db_name} AS "
            f"SELECT {full_select} "
            f"FROM {parquet_name} r "
            f"{join_clause}"
        )

    def _get_view_columns(self, view: str) -> list[str]:
        """
        Return column names for a view.

        Uses ``DESCRIBE`` rather than ``information_schema`` to force
        eager schema resolution for ``read_parquet``-backed views,
        which DuckDB may evaluate lazily.

        """
        df = self._conn.execute(f"DESCRIBE {view}").fetchdf()
        return df["column_name"].tolist()

    def _get_sample_id_col(self, db_name: str) -> str:
        """
        Resolve the sample identifier column name for a dataset.

        :param db_name: Resolved database view name
        :return: Actual column name for the sample identifier

        """
        repo_id, config_name = self.db_name_map[db_name]
        return self.config.get_sample_id_field(repo_id, config_name)

    def _resolve_metadata_fields(
        self, repo_id: str, config_name: str
    ) -> list[str] | None:
        """
        Get metadata field names from the DataCard.

        Delegates to ``DataCard.get_metadata_fields()`` which handles
        both embedded metadata_fields and external metadata configs
        (via applies_to).

        :param repo_id: Repository ID
        :param config_name: Configuration name
        :return: List of metadata field names, or None if not found

        """
        try:
            card = self.datacards.get(repo_id) or _cached_datacard(
                repo_id, token=self.token
            )
            return card.get_metadata_fields(config_name)
        except Exception:
            logger.error(
                "Could not resolve metadata_fields for %s/%s",
                repo_id,
                config_name,
            )
        return None

    def _get_class_label_names(
        self, card: Any, config_name: str, field: str
    ) -> list[str]:
        """
        Return the ENUM levels for a field with class_label dtype.

        Looks up the FeatureInfo for ``field`` in the DataCard config and
        extracts the ``names`` list from its ``class_label`` dtype dict.

        :param card: DataCard instance
        :param config_name: Configuration name
        :param field: Field name to look up
        :return: List of level strings
        :raises ValueError: If the field is not found, has no class_label dtype,
            or the class_label dict has no ``names`` key

        """
        try:
            features = card.get_features(config_name)
        except Exception as exc:
            raise ValueError(
                f"Could not retrieve features for config '{config_name}': {exc}"
            ) from exc

        feature = next((f for f in features if f.name == field), None)
        if feature is None:
            raise ValueError(
                f"Field '{field}' not found in DataCard config '{config_name}'. "
                "dtype='factor' requires the field to be declared in the DataCard."
            )

        dtype = feature.dtype
        if not isinstance(dtype, dict) or "class_label" not in dtype:
            raise ValueError(
                f"dtype='factor' is set for field '{field}' in config "
                f"'{config_name}', but the DataCard dtype is {dtype!r} rather "
                "than a class_label dict. "
                "The DataCard must declare dtype: {class_label: {names: [...]}}."
            )

        class_label = dtype["class_label"]
        names = class_label.get("names") if isinstance(class_label, dict) else None
        if not names:
            raise ValueError(
                f"class_label for field '{field}' in config '{config_name}' "
                "has no 'names' key or the names list is empty. "
                "Specify levels as: class_label: {names: [level1, level2, ...]}."
            )

        return [str(n) for n in names]

    def _ensure_enum_type(self, type_name: str, levels: list[str]) -> None:
        """
        Create or replace a DuckDB ENUM type with the given levels.

        DuckDB ENUM types must be registered before use in CAST expressions. Drops any
        existing type with the same name first to allow re-registration on repeated view
        builds.

        :param type_name: SQL identifier for the ENUM type
        :param levels: Ordered list of allowed string values

        """
        try:
            self._conn.execute(f"DROP TYPE IF EXISTS {type_name}")
        except Exception:
            pass  # type may not exist yet
        escaped = ", ".join(f"'{v.replace(chr(39), chr(39)*2)}'" for v in levels)
        self._conn.execute(f"CREATE TYPE {type_name} AS ENUM ({escaped})")

    def _resolve_alias(self, col: str, value: str) -> str:
        """
        Apply factor alias to a value if one is configured.

        :param col: Column name (e.g., "carbon_source")
        :param value: Raw value (e.g., "D-glucose")
        :return: Canonical alias (e.g., "glucose") or original value

        """
        aliases = self.config.factor_aliases.get(col)
        if not aliases:
            return value
        lower_val = str(value).lower()
        for canonical, actuals in aliases.items():
            if lower_val in [str(a).lower() for a in actuals]:
                return canonical
        return value

    def _resolve_property_columns(
        self,
        repo_id: str,
        config_name: str,
    ) -> tuple[list[str], list[str]] | None:
        """
        Build SQL column expressions for derived property columns.

        Resolves config property mappings against the DataCard to
        produce SQL expressions that add derived columns to the
        ``_meta`` view.

        :param repo_id: Repository ID
        :param config_name: Configuration name
        :return: Tuple of (sql_expressions, raw_cols_needed) or None
            if no property mappings are configured.
            ``sql_expressions`` are SQL fragments like
            ``"'glucose' AS carbon_source"`` or
            ``"CASE WHEN ... END AS carbon_source"``.
            ``raw_cols_needed`` are raw parquet column names that must
            be present in the inner SELECT.

        """
        mappings = self.config.get_property_mappings(repo_id, config_name)
        if not mappings and not self.config.missing_value_labels:
            return None

        expressions: list[str] = []
        raw_cols: set[str] = set()

        card = None
        if mappings:
            try:
                card = self.datacards.get(repo_id) or _cached_datacard(
                    repo_id, token=self.token
                )
            except Exception as exc:
                logger.warning(
                    "Could not load DataCard for %s: %s",
                    repo_id,
                    exc,
                )

        for key, mapping in mappings.items():
            if card is None:
                # Cannot resolve field/path mappings without a DataCard;
                # skip this mapping and fall through to missing_value_labels.
                continue
            if mapping.expression is not None:
                # Type D: expression
                expressions.append(f"({mapping.expression}) AS {_quote_ident(key)}")
                continue

            if mapping.field is not None and mapping.path is None:
                # Type A: field-only (alias or ENUM cast)
                raw_cols.add(mapping.field)
                if mapping.dtype == "factor":
                    # Fetch class_label levels from DataCard, register ENUM,
                    # and emit a CAST expression. Raises ValueError if the
                    # DataCard does not declare a class_label dtype.
                    enum_type = f"_enum_{key}"
                    levels = self._get_class_label_names(
                        card, config_name, mapping.field
                    )
                    self._ensure_enum_type(enum_type, levels)
                    # If all levels are integer-valued strings (e.g. '0',
                    # '90'), the parquet column may be DOUBLE (e.g. 90.0).
                    # Cast through BIGINT first to strip the decimal before
                    # converting to VARCHAR so '90.0' becomes '90'.
                    all_int = all(re.fullmatch(r"-?\d+", lv) for lv in levels)
                    inner = (
                        f"CAST({mapping.field} AS BIGINT)" if all_int else mapping.field
                    )
                    expressions.append(
                        f"CAST(CAST({inner} AS VARCHAR)"
                        f" AS {enum_type}) AS {_quote_ident(key)}"
                    )
                elif key == mapping.field:
                    # no-op -- column already present as raw col
                    pass
                else:
                    expressions.append(f"{mapping.field} AS {_quote_ident(key)}")
                continue

            if mapping.field is not None and mapping.path is not None:
                # Type B: field + path -- resolve from definitions.
                # dtype='factor' is not supported here: levels come from a
                # class_label field, not a definitions path.
                if mapping.dtype == "factor":
                    raise ValueError(
                        f"dtype='factor' is not supported for field+path mappings "
                        f"(key='{key}'). Use dtype='factor' only with field-only "
                        "mappings that reference a class_label field in the DataCard."
                    )
                raw_cols.add(mapping.field)
                expr = self._build_field_path_expr(
                    key,
                    mapping.field,
                    mapping.path,
                    mapping.dtype,
                    config_name,
                    card,
                )
                if expr is not None:
                    expressions.append(expr)
                continue

            if mapping.field is None and mapping.path is not None:
                # Type C: path-only -- constant from config
                expr = self._build_path_only_expr(
                    key,
                    mapping.path,
                    mapping.dtype,
                    config_name,
                    card,
                )
                if expr is not None:
                    expressions.append(expr)
                continue

        # For any key in missing_value_labels that was not covered by an
        # explicit mapping for this dataset, emit a constant literal so that
        # every _meta view exposes the column (with the fallback value).
        for key, label in self.config.missing_value_labels.items():
            if key not in mappings:
                escaped = label.replace("'", "''")
                expressions.append(f"'{escaped}' AS {_quote_ident(key)}")

        if not expressions and not raw_cols:
            return None

        return expressions, sorted(raw_cols)

    def _build_field_path_expr(
        self,
        key: str,
        field: str,
        path: str,
        dtype: str | None,
        config_name: str,
        card: Any,
    ) -> str | None:
        """
        Build a SQL expression for a field+path property mapping.

        Resolves each definition value via ``get_nested_value``,
        applies factor aliases, and returns either a constant or
        a CASE WHEN expression.

        :param key: Output column name
        :param field: Source field in parquet (e.g., "condition")
        :param path: Dot-notation path within definitions
        :param dtype: Optional data type ("numeric", "string", "bool")
        :param config_name: Configuration name
        :param card: DataCard instance
        :return: SQL expression string, or None on failure

        """
        try:
            defs = card.get_field_definitions(config_name, field)
        except Exception as exc:
            logger.warning(
                "Could not get definitions for field '%s' " "in config '%s': %s",
                field,
                config_name,
                exc,
            )
            return None

        if not defs:
            return None

        # Resolve each definition value
        value_map: dict[str, str] = {}
        for def_key, definition in defs.items():
            raw = get_nested_value(definition, path)
            if raw is None:
                logger.debug(
                    "Path '%s' resolved to None for " "definition key '%s' (keys: %s)",
                    path,
                    def_key,
                    (
                        list(definition.keys())
                        if isinstance(definition, dict)
                        else type(definition).__name__
                    ),
                )
                continue
            # Handle list results (e.g., carbon_source returns
            # [{"compound": "D-glucose"}])
            if isinstance(raw, list):
                raw = raw[0] if len(raw) == 1 else ", ".join(str(v) for v in raw)
            resolved = self._resolve_alias(key, str(raw))
            value_map[str(def_key)] = resolved

        if not value_map:
            return None

        # If all values are the same, emit a constant
        unique_vals = set(value_map.values())
        if len(unique_vals) == 1:
            val = next(iter(unique_vals))
            return self._literal_expr(key, val, dtype)

        # Otherwise, build CASE WHEN
        whens = []
        for def_key, resolved in value_map.items():
            escaped_key = def_key.replace("'", "''")
            escaped_val = resolved.replace("'", "''")
            whens.append(f"WHEN {field} = '{escaped_key}' " f"THEN '{escaped_val}'")
        case_sql = " ".join(whens)
        missing = self.config.missing_value_labels.get(key)
        if missing is not None:
            escaped_missing = missing.replace("'", "''")
            expr = f"CASE {case_sql} " f"ELSE '{escaped_missing}' END"
        else:
            expr = f"CASE {case_sql} ELSE NULL END"
        if dtype == "numeric":
            expr = f"CAST({expr} AS DOUBLE)"
        return f"{expr} AS {_quote_ident(key)}"

    def _build_path_only_expr(
        self,
        key: str,
        path: str,
        dtype: str | None,
        config_name: str,
        card: Any,
    ) -> str | None:
        """
        Build a constant column expression for a path-only mapping.

        Resolves a single value from the DataCard's raw model_extra,
        which preserves the full dict structure (including any
        ``experimental_conditions`` wrapper).

        :param key: Output column name
        :param path: Dot-notation path (may include
            ``experimental_conditions.`` prefix)
        :param dtype: Optional data type
        :param config_name: Configuration name
        :param card: DataCard instance
        :return: SQL literal expression, or None on failure

        """
        # Build merged dict from top-level + config-level model_extra.
        # This preserves keys like "experimental_conditions" that
        # get_experimental_conditions() would strip.
        merged: dict[str, Any] = {}
        try:
            top_extra = card.dataset_card.model_extra
            if isinstance(top_extra, dict):
                merged.update(top_extra)
            config_obj = card.get_config(config_name)
            if config_obj and isinstance(config_obj.model_extra, dict):
                merged.update(config_obj.model_extra)
        except Exception:
            logger.debug(
                "Could not get model_extra for %s/%s",
                card.repo_id if hasattr(card, "repo_id") else "?",
                config_name,
            )
            return None

        if not merged:
            return None

        raw = get_nested_value(merged, path)
        if raw is None:
            logger.debug(
                "Path '%s' resolved to None in model_extra for "
                "%s/%s. Available keys: %s",
                path,
                card.repo_id if hasattr(card, "repo_id") else "?",
                config_name,
                list(merged.keys()),
            )
            return None

        if isinstance(raw, list):
            raw = raw[0] if len(raw) == 1 else ", ".join(str(v) for v in raw)

        resolved = self._resolve_alias(key, str(raw))
        return self._literal_expr(key, resolved, dtype)

    @staticmethod
    def _literal_expr(key: str, value: str, dtype: str | None) -> str:
        """
        Build a SQL literal expression with optional type cast.

        :param key: Column alias
        :param value: Literal value
        :param dtype: Optional type ("numeric", "string", "bool")
        :return: SQL expression

        """
        escaped = value.replace("'", "''")
        if dtype == "numeric":
            return f"CAST('{escaped}' AS DOUBLE) AS {_quote_ident(key)}"
        return f"'{escaped}' AS {_quote_ident(key)}"

    def _register_comparative_expanded_view(
        self,
        db_name: str,
        ds_cfg: Any,
    ) -> None:
        """
        Create ``<db_name>_expanded`` view with parsed composite ID cols.

        For each link_field in the dataset config, adds two columns:

        - ``<link_field>_source`` -- the ``repo_id;config_name`` prefix,
          aliased to the configured ``db_name`` when available.
        - ``<link_field>_id`` -- the sample identifier component.

        :param db_name: Base view name for the comparative dataset
        :param ds_cfg: DatasetVirtualDBConfig with ``links``

        """
        parquet_view = f"__{db_name}_parquet"
        if not self._view_exists(parquet_view):
            return

        extra_cols = []
        for link_field, primaries in ds_cfg.links.items():
            # _id column: third component of composite ID
            id_col = f"{link_field}_id"
            extra_cols.append(f"SPLIT_PART({link_field}, ';', 3) " f"AS {id_col}")

            # _source column: first two components, aliased
            # to db_name when the pair is in the config
            raw_expr = (
                f"SPLIT_PART({link_field}, ';', 1) || ';' "
                f"|| SPLIT_PART({link_field}, ';', 2)"
            )
            whens = []
            for pair in primaries:
                repo_id, config_name = pair[0], pair[1]
                alias = self._get_db_name_for(repo_id, config_name)
                if alias:
                    key = f"{repo_id};{config_name}".replace("'", "''")
                    whens.append(f"WHEN '{key}' THEN '{alias}'")
            if whens:
                case_sql = " ".join(whens)
                source_expr = f"CASE {raw_expr} {case_sql} " f"ELSE {raw_expr} END"
            else:
                source_expr = raw_expr
            source_col = f"{link_field}_source"
            extra_cols.append(f"{source_expr} AS {source_col}")

        if not extra_cols:
            return

        cols_sql = ", ".join(extra_cols)
        self._conn.execute(
            f"CREATE OR REPLACE VIEW {db_name}_expanded AS "
            f"SELECT *, {cols_sql} FROM {parquet_view}"
        )

    # ------------------------------------------------------------------
    # Internal helpers
    # ------------------------------------------------------------------

    def _is_comparative(self, repo_id: str, config_name: str) -> bool:
        """Return True if the dataset has links (i.e. is comparative)."""
        repo_cfg = self.config.repositories.get(repo_id)
        if not repo_cfg or not repo_cfg.dataset:
            return False
        ds_cfg = repo_cfg.dataset.get(config_name)
        return bool(ds_cfg and ds_cfg.links)

    def _list_views(self) -> list[str]:
        """Return list of public views (excludes internal __ prefixed)."""
        df = self._conn.execute(
            "SELECT table_name FROM information_schema.tables "
            "WHERE table_schema = 'main' AND table_type = 'VIEW'"
        ).fetchdf()
        return [n for n in df["table_name"].tolist() if not n.startswith("__")]

    def _view_exists(self, name: str) -> bool:
        """Check whether a view is registered (including internal)."""
        df = self._conn.execute(
            "SELECT table_name FROM information_schema.tables "
            "WHERE table_schema = 'main' AND table_type = 'VIEW' "
            f"AND table_name = '{name}'"
        ).fetchdf()
        return len(df) > 0

    def _get_primary_view_names(self) -> list[str]:
        """
        Return db_names of primary (non-comparative) raw views.

        A primary dataset is one whose config has no ``links``.

        """
        names = []
        for db_name, (repo_id, config_name) in self.db_name_map.items():
            if not self._is_comparative(repo_id, config_name):
                if self._view_exists(db_name):
                    names.append(db_name)
        return sorted(names)

    def _get_primary_meta_view_names(self) -> list[str]:
        """Return names of primary ``_meta`` views."""
        return [
            f"{n}_meta"
            for n in self._get_primary_view_names()
            if self._view_exists(f"{n}_meta")
        ]

    def _get_db_name_for(self, repo_id: str, config_name: str) -> str | None:
        """Resolve db_name for a (repo_id, config_name) pair."""
        for db_name, (r, c) in self.db_name_map.items():
            if r == repo_id and c == config_name:
                return db_name
        return None

    def __repr__(self) -> str:
        """String representation."""
        n_repos = len(self.config.repositories)
        n_datasets = len(self.db_name_map)
        n_views = len(self._list_views())
        return (
            f"VirtualDB({n_repos} repos, "
            f"{n_datasets} datasets, "
            f"{n_views} views)"
        )

__init__(config_path, token=None, duckdb_connection=None, local_files_only=False, cache_dir=None)

Initialize VirtualDB with configuration.

Creates the DuckDB connection and registers all views immediately.

Parameters:

Name Type Description Default
config_path Path | str

Path to YAML configuration file

required
token str | None

Optional HuggingFace token for private datasets

None
duckdb_connection DuckDBPyConnection | None

Optional DuckDB connection. If provided, views will be registered on this connection instead of creating a new in-memory database. This provides a method of using a persistent database file. If not provided, an in-memory DuckDB connection is created.

None
local_files_only bool

If True, skip HuggingFace network checks and use only locally cached files. Eliminates per-dataset repo_info HTTP round-trips on warm restarts. Raises huggingface_hub.utils.LocalEntryNotFoundError if any required file is absent from the local cache.

False
cache_dir Path | str | None

Override the HuggingFace cache directory. When None, huggingface_hub resolves the location from HF_CACHE_DIR / HF_HOME / its own default. Pass an explicit path to store or read parquet snapshots from a non-default location (e.g. a bundled directory inside the deployed application).

None

Raises:

Type Description
FileNotFoundError

If config file does not exist

ValueError

If configuration is invalid

Source code in labretriever/virtual_db.py
def __init__(
    self,
    config_path: Path | str,
    token: str | None = None,
    duckdb_connection: duckdb.DuckDBPyConnection | None = None,
    local_files_only: bool = False,
    cache_dir: Path | str | None = None,
):
    """
    Initialize VirtualDB with configuration.

    Creates the DuckDB connection and registers all views immediately.

    :param config_path: Path to YAML configuration file
    :param token: Optional HuggingFace token for private datasets
    :param duckdb_connection: Optional DuckDB connection. If provided, views will be
        registered on this connection instead of creating a new in-memory database.
        This provides a method of using a persistent database file. If not provided,
        an in-memory DuckDB connection is created.
    :param local_files_only: If ``True``, skip HuggingFace network checks and use
        only locally cached files. Eliminates per-dataset ``repo_info``
        HTTP round-trips on warm restarts. Raises
        ``huggingface_hub.utils.LocalEntryNotFoundError`` if
        any required file is absent from the local cache.
    :param cache_dir: Override the HuggingFace cache directory. When ``None``,
        ``huggingface_hub`` resolves the location from ``HF_CACHE_DIR`` /
        ``HF_HOME`` / its own default. Pass an explicit path to store or read
        parquet snapshots from a non-default location (e.g. a bundled directory
        inside the deployed application).
    :raises FileNotFoundError: If config file does not exist
    :raises ValueError: If configuration is invalid

    """
    self.config = MetadataConfig.from_yaml(config_path)
    self.token = token
    self.local_files_only = local_files_only
    # Explicit argument wins; fall back to get_cache_dir() which reads
    # HF_CACHE_DIR at call time so CLI --cache-dir is respected even when
    # VirtualDB is constructed without the parameter.
    from labretriever.constants import get_cache_dir

    self.cache_dir: Path = (
        Path(cache_dir) if cache_dir is not None else get_cache_dir()
    )

    self._conn: duckdb.DuckDBPyConnection = (
        duckdb_connection
        if duckdb_connection is not None
        else duckdb.connect(":memory:")
    )

    # db_name -> (repo_id, config_name)
    self.db_name_map = self._build_db_name_map()

    # Prepared queries: name -> sql
    self._prepared_queries: dict[str, str] = {}

    _t0 = time.monotonic()

    t = time.monotonic()
    self._load_datacards()
    logger.debug("_load_datacards completed in %.3fs", time.monotonic() - t)

    t = time.monotonic()
    self._validate_datacards()
    logger.debug("_validate_datacards completed in %.3fs", time.monotonic() - t)

    t = time.monotonic()
    self._update_cache()
    logger.debug("_update_cache completed in %.3fs", time.monotonic() - t)

    t = time.monotonic()
    self._register_all_views()
    logger.debug("_register_all_views completed in %.3fs", time.monotonic() - t)

    t = time.monotonic()
    self._build_column_metadata()
    logger.debug("_build_column_metadata completed in %.3fs", time.monotonic() - t)

    logger.debug("VirtualDB.__init__ total: %.3fs", time.monotonic() - _t0)

__repr__()

String representation.

Source code in labretriever/virtual_db.py
def __repr__(self) -> str:
    """String representation."""
    n_repos = len(self.config.repositories)
    n_datasets = len(self.db_name_map)
    n_views = len(self._list_views())
    return (
        f"VirtualDB({n_repos} repos, "
        f"{n_datasets} datasets, "
        f"{n_views} views)"
    )

describe(table=None)

Describe column names and types for one or all views.

Parameters:

Name Type Description Default
table str | None

View name, or None for all views

None

Returns:

Type Description
DataFrame

DataFrame with columns table, column_name, column_type

Source code in labretriever/virtual_db.py
def describe(self, table: str | None = None) -> pd.DataFrame:
    """
    Describe column names and types for one or all views.

    :param table: View name, or None for all views
    :return: DataFrame with columns ``table``, ``column_name``,
             ``column_type``

    """

    if table is not None:
        df = self._conn.execute(f"DESCRIBE {table}").fetchdf()
        df.insert(0, "table", table)
        return df

    frames = []
    for view in sorted(self._list_views()):
        df = self._conn.execute(f"DESCRIBE {view}").fetchdf()
        df.insert(0, "table", view)
        frames.append(df)
    if not frames:
        return pd.DataFrame(columns=["table", "column_name", "column_type"])
    return pd.concat(frames, ignore_index=True)

get_citation(db_name)

Return the citation for a dataset.

The dataset-level citation takes precedence over the repository-level citation. If neither is defined, returns None.

Parameters:

Name Type Description Default
db_name str

Dataset name as registered in the VirtualDB config.

required

Returns:

Type Description
str | None

Citation string, or None if no citation is defined.

Source code in labretriever/virtual_db.py
def get_citation(self, db_name: str) -> str | None:
    """
    Return the citation for a dataset.

    The dataset-level citation takes precedence over the repository-level
    citation. If neither is defined, returns ``None``.

    :param db_name: Dataset name as registered in the VirtualDB config.
    :returns: Citation string, or ``None`` if no citation is defined.
    :rtype: str | None

    """
    if db_name not in self.db_name_map:
        return None
    repo_id, config_name = self.db_name_map[db_name]
    card = self.datacards.get(repo_id)
    if card is None:
        return None
    return card.get_citation(config_name)

get_column_metadata(db_name)

Return per-column metadata for a primary dataset.

Metadata is collected from the DataCard during VirtualDB construction and includes the feature description, semantic role, and per-level definitions for experimental_condition columns.

Parameters:

Name Type Description Default
db_name str

Dataset name as registered in the VirtualDB config.

required

Returns:

Type Description
dict[str, ColumnMeta] | None

Dict mapping column name to :class:ColumnMeta, or None if db_name is not found or has no recorded metadata.

Source code in labretriever/virtual_db.py
def get_column_metadata(self, db_name: str) -> dict[str, ColumnMeta] | None:
    """
    Return per-column metadata for a primary dataset.

    Metadata is collected from the DataCard during VirtualDB construction
    and includes the feature description, semantic role, and per-level
    definitions for ``experimental_condition`` columns.

    :param db_name: Dataset name as registered in the VirtualDB config.
    :returns: Dict mapping column name to :class:`ColumnMeta`, or ``None``
        if ``db_name`` is not found or has no recorded metadata.
    :rtype: dict[str, ColumnMeta] | None

    """
    return self._column_metadata.get(db_name)

get_common_fields()

Return columns present in ALL primary _meta views.

Primary dataset views are those without links in their config (i.e. not comparative datasets).

Returns:

Type Description
list[str]

Sorted list of common column names

Source code in labretriever/virtual_db.py
def get_common_fields(self) -> list[str]:
    """
    Return columns present in ALL primary ``_meta`` views.

    Primary dataset views are those without ``links`` in their
    config (i.e. not comparative datasets).

    :return: Sorted list of common column names

    """

    meta_views = self._get_primary_meta_view_names()
    if not meta_views:
        return []

    sets = []
    for view in meta_views:
        cols = self._conn.execute(
            f"SELECT column_name FROM information_schema.columns "
            f"WHERE table_name = '{view}'"
        ).fetchdf()
        sets.append(set(cols["column_name"].tolist()))

    common = set.intersection(*sets)
    return sorted(common)

get_condition_field_info(db_name)

Return hierarchically linked property column groups for a dataset.

Some datasets expose multiple derived property columns in their _meta view (e.g. "Carbon source" and "Temperature") that are both extracted from the same per-sample experimental-condition field via field+path PropertyMapping entries in the VirtualDB configuration. Because these columns are co-derived from the same source field, the values that can appear in one column are constrained by the value selected in another (selecting Carbon source = glucose limits which Temperature values are meaningful).

This method identifies such groups and enriches them with per-level descriptions from the DataCard, making it straightforward for downstream tools to build conditional filter interfaces.

Parameters:

Name Type Description Default
db_name str

Dataset name as returned by :meth:get_datasets.

required

Returns:

Type Description
dict[str, Any] | None Example:: info = vdb.get_condition_field_info("harbison") if info is not None: for src_field, group in info.items(): print(f"Source field '{src_field}' links:") print(f" property columns: {group['property_cols']}") for level, desc in group["level_descriptions"].items(): print(f" {level}: {desc}")

A dict keyed by source parquet field name. Each value is a dict with two keys:

"property_cols"list[str] The derived property column names that appear in the {db_name}_meta view and share this source field.

"level_descriptions"dict[str, str] Maps each categorical level of the source field to a human-readable description string retrieved from the DataCard via :meth:~labretriever.datacard.DataCard.get_field_definitions. # noqa: E501 Levels without a "description" key in their definition dict default to "Description unavailable".

Returns None when the dataset has no linked property groups (no Type-A field-only mapping exists, or there are no other property columns to serve as downstream members), or when db_name is not found.

Source code in labretriever/virtual_db.py
def get_condition_field_info(self, db_name: str) -> dict[str, Any] | None:
    """
    Return hierarchically linked property column groups for a dataset.

    Some datasets expose multiple derived property columns in their
    ``_meta`` view (e.g. ``"Carbon source"`` and ``"Temperature"``) that
    are both extracted from the same per-sample experimental-condition
    field via field+path ``PropertyMapping`` entries in the VirtualDB
    configuration.  Because these columns are co-derived from the same
    source field, the values that can appear in one column are constrained
    by the value selected in another (selecting ``Carbon source =
    glucose`` limits which ``Temperature`` values are meaningful).

    This method identifies such groups and enriches them with per-level
    descriptions from the DataCard, making it straightforward for
    downstream tools to build conditional filter interfaces.

    :param db_name: Dataset name as returned by :meth:`get_datasets`.
    :type db_name: str
    :returns: A dict keyed by source parquet field name.  Each value is
        a dict with two keys:

        ``"property_cols"`` — ``list[str]``
            The derived property column names that appear in the
            ``{db_name}_meta`` view and share this source field.

        ``"level_descriptions"`` — ``dict[str, str]``
            Maps each categorical level of the source field to a
            human-readable description string retrieved from the
            DataCard via :meth:`~labretriever.datacard.DataCard.get_field_definitions`. # noqa: E501
            Levels without a ``"description"`` key in their definition
            dict default to ``"Description unavailable"``.

        Returns ``None`` when the dataset has no linked property groups
        (no Type-A field-only mapping exists, or there are no other
        property columns to serve as downstream members),
        or when ``db_name`` is not found.
    :rtype: dict[str, Any] | None

    Example::

        info = vdb.get_condition_field_info("harbison")
        if info is not None:
            for src_field, group in info.items():
                print(f"Source field '{src_field}' links:")
                print(f"  property columns: {group['property_cols']}")
                for level, desc in group["level_descriptions"].items():
                    print(f"  {level}: {desc}")

    """
    warnings.warn(
        "get_condition_field_info is deprecated; use get_column_metadata instead.",
        DeprecationWarning,
        stacklevel=2,
    )
    if db_name not in self.db_name_map:
        return None

    repo_id, config_name = self.db_name_map[db_name]
    mappings = self.config.get_property_mappings(repo_id, config_name)

    # Collect Type-C (path-only) columns — repo-level experimental condition
    # extractions.  These are shared downstream members for any Type-A anchor
    # that has at least one co-occurring Type-B sibling or Type-C companion.
    type_c_cols = [
        prop_col
        for prop_col, pm in mappings.items()
        if pm.field is None and pm.path is not None
    ]

    # Build per-src_field groups anchored by Type-A (field-only) mappings.
    # A Type-A mapping is a condition anchor only when its prop_col name
    # differs from pm.field — this distinguishes semantic aliases like
    # "Experimental condition" (field=condition) from identity renames like
    # "regulator_locus_tag" (field=regulator_locus_tag).
    # The group requires at least one downstream member: a Type-B col sharing
    # the same src_field, or a Type-C col (when no Type-B siblings exist).
    linked: dict[str, dict[str, list[str]]] = {}
    for prop_col, pm in mappings.items():
        if pm.field is None or pm.path is not None:
            continue  # not Type-A
        # Exclude identity renames: prop_col and pm.field are the same name,
        # possibly differing only in case or whitespace (e.g. "Regulator locus tag"
        # -> "regulator_locus_tag").  Normalise both sides before comparing.
        if prop_col.lower().replace(" ", "_") == pm.field.lower().replace(" ", "_"):
            continue
        src_field = pm.field
        type_b_for_field = [
            col
            for col, m in mappings.items()
            if m.field == src_field and m.path is not None
        ]
        type_c_for_group = [c for c in type_c_cols if c not in type_b_for_field]
        downstream = type_b_for_field + type_c_for_group
        if downstream:
            linked[src_field] = {
                "prop_cols": downstream,
                "type_b_cols": type_b_for_field,
            }

    card = self.datacards.get(repo_id)

    # Fallback: when no Type-A aliases exist but there are Type-C downstream
    # cols, use the DataCard's condition_fields (role=experimental_condition)
    # as implicit anchors.  Use the first condition field only; multiple
    # condition fields sharing identical downstream cols would produce
    # duplicate input IDs in the UI.
    # Fallback: DataCard condition_fields as implicit anchors.  All Type-C
    # cols are downstream; none are Type-B (no per-level field derivation).
    if not linked and type_c_cols and card is not None:
        try:
            schema = card.extract_metadata_schema(config_name)
            cond_fields = schema.get("condition_fields", [])
            if cond_fields:
                linked[cond_fields[0]] = {
                    "prop_cols": list(type_c_cols),
                    "type_b_cols": [],
                }
        except Exception:
            pass

    if not linked:
        return None

    # Resolve the config name to try for field definitions.  For datasets
    # with external metadata, the definitions live in the _meta config rather
    # than the primary annotated-features config, so prefer that when present.
    meta_config_name = self._external_meta_configs.get(db_name, config_name)

    result: dict[str, Any] = {}
    for src_field, col_info in linked.items():
        prop_cols = col_info["prop_cols"]
        type_b_cols = col_info["type_b_cols"]
        level_descriptions: dict[str, str] = {}
        if card is not None:
            for _cfg in (meta_config_name, config_name):
                try:
                    defs = card.get_field_definitions(_cfg, src_field)
                    if defs:
                        for level_val, definition in defs.items():
                            desc = (
                                definition.get(
                                    "description", "Description unavailable"
                                )
                                if isinstance(definition, dict)
                                else "Description unavailable"
                            )
                            level_descriptions[str(level_val)] = desc
                        break  # found definitions — stop trying configs
                except Exception:
                    logger.debug(
                        "No field definitions for %s/%s field '%s'",
                        db_name,
                        _cfg,
                        src_field,
                    )
        result[src_field] = {
            "property_cols": prop_cols,
            "type_b_cols": type_b_cols,
            "level_descriptions": level_descriptions,
        }
    return result

get_dataset_description(db_name)

Return the description for a dataset.

The VirtualDB config description field takes precedence over the DataCard config description. Returns None if neither is defined.

Parameters:

Name Type Description Default
db_name str

Dataset name as registered in the VirtualDB config.

required

Returns:

Type Description
str | None

Description string, or None if not found.

Source code in labretriever/virtual_db.py
def get_dataset_description(self, db_name: str) -> str | None:
    """
    Return the description for a dataset.

    The VirtualDB config ``description`` field takes precedence over the
    DataCard config description. Returns ``None`` if neither is defined.

    :param db_name: Dataset name as registered in the VirtualDB config.
    :returns: Description string, or ``None`` if not found.
    :rtype: str | None

    """
    if db_name not in self.db_name_map:
        return None
    repo_id, config_name = self.db_name_map[db_name]
    # Check VirtualDB config description override first
    repo_cfg = self.config.repositories.get(repo_id)
    if repo_cfg and repo_cfg.dataset:
        vdb_dataset_cfg = repo_cfg.dataset.get(config_name)
        if vdb_dataset_cfg and vdb_dataset_cfg.description is not None:
            return vdb_dataset_cfg.description
    # Fall back to DataCard description
    card = self.datacards.get(repo_id)
    if card is None:
        return None
    cfg = card.get_config(config_name)
    return cfg.description if cfg is not None else None

get_datasets()

Return the sorted list of dataset names known to this VirtualDB.

Dataset names are the resolved db_name values from the configuration (falling back to the config_name when db_name is not explicitly set). These are the names accepted by :meth:get_tags and queryable via :meth:query.

Unlike :meth:tables, this method reads directly from the configuration and does not require views to be registered, so no data is downloaded.

Returns:

Type Description
list[str]

Sorted list of dataset names

Source code in labretriever/virtual_db.py
def get_datasets(self) -> list[str]:
    """
    Return the sorted list of dataset names known to this VirtualDB.

    Dataset names are the resolved ``db_name`` values from the
    configuration (falling back to the config_name when ``db_name``
    is not explicitly set). These are the names accepted by
    :meth:`get_tags` and queryable via :meth:`query`.

    Unlike :meth:`tables`, this method reads directly from the
    configuration and does not require views to be registered, so
    no data is downloaded.

    :return: Sorted list of dataset names

    """
    return sorted(self.db_name_map)

get_fields(table=None)

Return column names for a view or all unique columns.

Parameters:

Name Type Description Default
table str | None

View name, or None for all views

None

Returns:

Type Description
list[str]

Sorted list of column names

Source code in labretriever/virtual_db.py
def get_fields(self, table: str | None = None) -> list[str]:
    """
    Return column names for a view or all unique columns.

    :param table: View name, or None for all views
    :return: Sorted list of column names

    """

    if table is not None:
        cols = self._conn.execute(
            f"SELECT column_name FROM information_schema.columns "
            f"WHERE table_name = '{table}'"
        ).fetchdf()
        return sorted(cols["column_name"].tolist())

    all_cols: set[str] = set()
    for view in self._list_views():
        cols = self._conn.execute(
            f"SELECT column_name FROM information_schema.columns "
            f"WHERE table_name = '{view}'"
        ).fetchdf()
        all_cols.update(cols["column_name"].tolist())
    return sorted(all_cols)

get_tags(db_name)

Return the merged tags for a dataset.

Tags are defined in the configuration at the repository and/or dataset level. Dataset-level tags override repository-level tags with the same key. See the tags section of the configuration guide for details.

Parameters:

Name Type Description Default
db_name str

Dataset name as it appears in :meth:tables (the resolved db_name from the configuration, or the config_name if db_name was not explicitly set).

required

Returns:

Type Description
dict[str, str]

Dict of merged tags, or empty dict if the dataset has no tags or the name is not found.

Source code in labretriever/virtual_db.py
def get_tags(self, db_name: str) -> dict[str, str]:
    """
    Return the merged tags for a dataset.

    Tags are defined in the configuration at the repository and/or
    dataset level. Dataset-level tags override repository-level tags
    with the same key. See the ``tags`` section of the configuration
    guide for details.

    :param db_name: Dataset name as it appears in :meth:`tables` (the
        resolved ``db_name`` from the configuration, or the
        ``config_name`` if ``db_name`` was not explicitly set).
    :return: Dict of merged tags, or empty dict if the dataset has no
        tags or the name is not found.

    """
    if db_name not in self.db_name_map:
        return {}
    repo_id, config_name = self.db_name_map[db_name]
    return self.config.get_tags(repo_id, config_name)

materialize()

Replace all registered dataset views with in-memory DuckDB tables.

For each dataset known to this VirtualDB, reads the public data view and the corresponding _meta view into in-memory tables, then re-points the original view names at those tables. Subsequent queries hit RAM instead of re-scanning parquet files, which substantially reduces per-query latency at the cost of increased startup time and memory usage.

Only views that exist in the current DuckDB session are materialized; missing views are silently skipped. Safe to call multiple times — uses CREATE OR REPLACE semantics.

Source code in labretriever/virtual_db.py
def materialize(self) -> None:
    """
    Replace all registered dataset views with in-memory DuckDB tables.

    For each dataset known to this VirtualDB, reads the public data view
    and the corresponding ``_meta`` view into in-memory tables, then
    re-points the original view names at those tables. Subsequent queries
    hit RAM instead of re-scanning parquet files, which substantially
    reduces per-query latency at the cost of increased startup time and
    memory usage.

    Only views that exist in the current DuckDB session are materialized;
    missing views are silently skipped. Safe to call multiple times —
    uses ``CREATE OR REPLACE`` semantics.

    """
    conn = self._conn
    for db_name in self.get_datasets():
        for view_name in (db_name, f"{db_name}_meta"):
            row = conn.execute(
                "SELECT view_name FROM duckdb_views() WHERE view_name = ?",
                [view_name],
            ).fetchone()
            if row is None:
                continue
            table_name = f"_mat_{view_name}"
            conn.execute(
                f"CREATE OR REPLACE TABLE {table_name} AS SELECT * FROM {view_name}"
            )
            conn.execute(
                f"CREATE OR REPLACE VIEW {view_name} AS SELECT * FROM {table_name}"
            )
            logger.debug("materialize: %s -> %s", view_name, table_name)

prepare(name, sql, overwrite=False)

Register a named parameterized query for later use.

Parameters use DuckDB $name syntax.

Parameters:

Name Type Description Default
name str

Query name (must not collide with a view name)

required
sql str

SQL template with $name parameters

required
overwrite bool

If True, overwrite existing prepared query with same name

False

Raises:

Type Description
ValueError

If name collides with an existing view

Example::

vdb.prepare("glucose_regs", '''
    SELECT regulator_symbol, COUNT(*) AS n
    FROM harbison_meta
    WHERE carbon_source = $cs
    GROUP BY regulator_symbol
    HAVING n >= $min_n
''')
df = vdb.query("glucose_regs", cs="glucose", min_n=2)
Source code in labretriever/virtual_db.py
def prepare(self, name: str, sql: str, overwrite: bool = False) -> None:
    """
    Register a named parameterized query for later use.

    Parameters use DuckDB ``$name`` syntax.

    :param name: Query name (must not collide with a view name)
    :param sql: SQL template with ``$name`` parameters
    :param overwrite: If True, overwrite existing prepared query
        with same name
    :raises ValueError: If *name* collides with an existing view

    Example::

        vdb.prepare("glucose_regs", '''
            SELECT regulator_symbol, COUNT(*) AS n
            FROM harbison_meta
            WHERE carbon_source = $cs
            GROUP BY regulator_symbol
            HAVING n >= $min_n
        ''')
        df = vdb.query("glucose_regs", cs="glucose", min_n=2)

    """

    if name in self._list_views() and not overwrite:
        error_msg = (
            f"Prepared-query name '{name}' collides with "
            f"an existing view. Choose a different name or set "
            f"overwrite=True."
        )
        logger.error(error_msg)
        raise ValueError(error_msg)
    self._prepared_queries[name] = sql

query(sql, **params)

Execute SQL or a prepared query and return a DataFrame.

If sql matches a registered prepared-query name the stored SQL template is used instead. Keyword arguments are passed as named parameters to DuckDB.

Parameters:

Name Type Description Default
sql str

Raw SQL string or name of a prepared query

required
params Any

Named parameters (DuckDB $name syntax)

{}

Returns:

Type Description
DataFrame

Query result as a pandas DataFrame

Examples::

# Raw SQL
df = vdb.query("SELECT * FROM harbison LIMIT 5")

# With parameters
df = vdb.query(
    "SELECT * FROM harbison_meta WHERE carbon_source = $cs",
    cs="glucose",
)

# Prepared query
vdb.prepare("top", "SELECT * FROM harbison_meta LIMIT $n")
df = vdb.query("top", n=10)
Source code in labretriever/virtual_db.py
def query(self, sql: str, **params: Any) -> pd.DataFrame:
    """
    Execute SQL or a prepared query and return a DataFrame.

    If *sql* matches a registered prepared-query name the stored
    SQL template is used instead. Keyword arguments are passed as
    named parameters to DuckDB.

    :param sql: Raw SQL string **or** name of a prepared query
    :param params: Named parameters (DuckDB ``$name`` syntax)
    :return: Query result as a pandas DataFrame

    Examples::

        # Raw SQL
        df = vdb.query("SELECT * FROM harbison LIMIT 5")

        # With parameters
        df = vdb.query(
            "SELECT * FROM harbison_meta WHERE carbon_source = $cs",
            cs="glucose",
        )

        # Prepared query
        vdb.prepare("top", "SELECT * FROM harbison_meta LIMIT $n")
        df = vdb.query("top", n=10)

    """
    # param `sql` may be a prepared query name, a raw sql statement, or
    # a parameterized sql statement that is not prepared. If it exists as a key
    # in the _prepared_queries dict, we use the prepared sql. Otherwise, we
    # use the sql as passed to query().
    resolved = self._prepared_queries.get(sql, sql)
    try:
        if params:
            return self._conn.execute(resolved, params).fetchdf()
        return self._conn.execute(resolved).fetchdf()
    except Exception as exc:
        import pprint

        params_repr = pprint.pformat(params, indent=2)
        raise QueryError(
            f"query failed: {exc}\n\n" f"SQL:\n{sql}\n\n" f"params:\n{params_repr}"
        ) from exc

tables()

Return sorted list of registered view names.

Returns:

Type Description
list[str]

Sorted list of view names

Source code in labretriever/virtual_db.py
def tables(self) -> list[str]:
    """
    Return sorted list of registered view names.

    :return: Sorted list of view names

    """

    return sorted(self._list_views())