File size: 95,878 Bytes
9c6594c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 |
import copy
import os
from collections.abc import Iterator
from functools import partial
from itertools import groupby
from typing import TYPE_CHECKING, Any, Callable, Optional, TypeVar, Union
import numpy as np
import pyarrow as pa
import pyarrow.compute as pc
import pyarrow.types
from .utils.logging import get_logger
if TYPE_CHECKING:
from .features.features import Features, FeatureType
logger = get_logger(__name__)
def inject_arrow_table_documentation(arrow_table_method):
def wrapper(fn):
fn.__doc__ = arrow_table_method.__doc__ + (fn.__doc__ if fn.__doc__ is not None else "")
fn.__doc__ = fn.__doc__.replace("pyarrow.Table", "Table")
if hasattr(arrow_table_method, "__annotations__"):
fn.__annotations__ = arrow_table_method.__annotations__
return fn
return wrapper
def _in_memory_arrow_table_from_file(filename: str) -> pa.Table:
in_memory_stream = pa.input_stream(filename)
opened_stream = pa.ipc.open_stream(in_memory_stream)
pa_table = opened_stream.read_all()
return pa_table
def _in_memory_arrow_table_from_buffer(buffer: pa.Buffer) -> pa.Table:
stream = pa.BufferReader(buffer)
opened_stream = pa.ipc.open_stream(stream)
table = opened_stream.read_all()
return table
def _memory_mapped_record_batch_reader_from_file(filename: str) -> pa.RecordBatchStreamReader:
memory_mapped_stream = pa.memory_map(filename)
return pa.ipc.open_stream(memory_mapped_stream)
def read_schema_from_file(filename: str) -> pa.Schema:
"""
Infer arrow table schema from file without loading whole file into memory.
Useful especially while having very big files.
"""
with pa.memory_map(filename) as memory_mapped_stream:
schema = pa.ipc.open_stream(memory_mapped_stream).schema
return schema
def _memory_mapped_arrow_table_from_file(filename: str) -> pa.Table:
opened_stream = _memory_mapped_record_batch_reader_from_file(filename)
pa_table = opened_stream.read_all()
return pa_table
def _deepcopy(x, memo: dict):
"""deepcopy a regular class instance"""
cls = x.__class__
result = cls.__new__(cls)
memo[id(x)] = result
for k, v in x.__dict__.items():
setattr(result, k, copy.deepcopy(v, memo))
return result
def _interpolation_search(arr: list[int], x: int) -> int:
"""
Return the position i of a sorted array so that arr[i] <= x < arr[i+1]
Args:
arr (`List[int]`): non-empty sorted list of integers
x (`int`): query
Returns:
`int`: the position i so that arr[i] <= x < arr[i+1]
Raises:
`IndexError`: if the array is empty or if the query is outside the array values
"""
i, j = 0, len(arr) - 1
while i < j and arr[i] <= x < arr[j]:
k = i + ((j - i) * (x - arr[i]) // (arr[j] - arr[i]))
if arr[k] <= x < arr[k + 1]:
return k
elif arr[k] < x:
i, j = k + 1, j
else:
i, j = i, k
raise IndexError(f"Invalid query '{x}' for size {arr[-1] if len(arr) else 'none'}.")
class IndexedTableMixin:
def __init__(self, table: pa.Table):
self._schema: pa.Schema = table.schema
self._batches: list[pa.RecordBatch] = [
recordbatch for recordbatch in table.to_batches() if len(recordbatch) > 0
]
self._offsets: np.ndarray = np.cumsum([0] + [len(b) for b in self._batches], dtype=np.int64)
def fast_gather(self, indices: Union[list[int], np.ndarray]) -> pa.Table:
"""
Create a pa.Table by gathering the records at the records at the specified indices. Should be faster
than pa.concat_tables(table.fast_slice(int(i) % table.num_rows, 1) for i in indices) since NumPy can compute
the binary searches in parallel, highly optimized C
"""
if not len(indices):
raise ValueError("Indices must be non-empty")
batch_indices = np.searchsorted(self._offsets, indices, side="right") - 1
return pa.Table.from_batches(
[
self._batches[batch_idx].slice(i - self._offsets[batch_idx], 1)
for batch_idx, i in zip(batch_indices, indices)
],
schema=self._schema,
)
def fast_slice(self, offset=0, length=None) -> pa.Table:
"""
Slice the Table using interpolation search.
The behavior is the same as `pyarrow.Table.slice` but it's significantly faster.
Interpolation search is used to find the start and end indexes of the batches we want to keep.
The batches to keep are then concatenated to form the sliced Table.
"""
if offset < 0:
raise IndexError("Offset must be non-negative")
elif offset >= self._offsets[-1] or (length is not None and length <= 0):
return pa.Table.from_batches([], schema=self._schema)
i = _interpolation_search(self._offsets, offset)
if length is None or length + offset >= self._offsets[-1]:
batches = self._batches[i:]
batches[0] = batches[0].slice(offset - self._offsets[i])
else:
j = _interpolation_search(self._offsets, offset + length - 1)
batches = self._batches[i : j + 1]
batches[-1] = batches[-1].slice(0, offset + length - self._offsets[j])
batches[0] = batches[0].slice(offset - self._offsets[i])
return pa.Table.from_batches(batches, schema=self._schema)
class Table(IndexedTableMixin):
"""
Wraps a pyarrow Table by using composition.
This is the base class for `InMemoryTable`, `MemoryMappedTable` and `ConcatenationTable`.
It implements all the basic attributes/methods of the pyarrow Table class except
the Table transforms: `slice, filter, flatten, combine_chunks, cast, add_column,
append_column, remove_column, set_column, rename_columns` and `drop`.
The implementation of these methods differs for the subclasses.
"""
def __init__(self, table: pa.Table):
super().__init__(table)
self.table = table
def __deepcopy__(self, memo: dict):
# arrow tables are immutable, so there's no need to copy self.table
# moreover calling deepcopy on a pyarrow table seems to make pa.total_allocated_bytes() decrease for some reason
# by adding it to the memo, self.table won't be copied
memo[id(self.table)] = self.table
# same for the recordbatches used by the index
memo[id(self._batches)] = list(self._batches)
return _deepcopy(self, memo)
def validate(self, *args, **kwargs):
"""
Perform validation checks. An exception is raised if validation fails.
By default only cheap validation checks are run. Pass `full=True`
for thorough validation checks (potentially `O(n)`).
Args:
full (`bool`, defaults to `False`):
If `True`, run expensive checks, otherwise cheap checks only.
Raises:
`pa.lib.ArrowInvalid`: if validation fails
"""
return self.table.validate(*args, **kwargs)
def equals(self, *args, **kwargs):
"""
Check if contents of two tables are equal.
Args:
other ([`~datasets.table.Table`]):
Table to compare against.
check_metadata `bool`, defaults to `False`):
Whether schema metadata equality should be checked as well.
Returns:
`bool`
"""
args = tuple(arg.table if isinstance(arg, Table) else arg for arg in args)
kwargs = {k: v.table if isinstance(v, Table) else v for k, v in kwargs}
return self.table.equals(*args, **kwargs)
def to_batches(self, *args, **kwargs):
"""
Convert Table to list of (contiguous) `RecordBatch` objects.
Args:
max_chunksize (`int`, defaults to `None`):
Maximum size for `RecordBatch` chunks. Individual chunks may be
smaller depending on the chunk layout of individual columns.
Returns:
`List[pyarrow.RecordBatch]`
"""
return self.table.to_batches(*args, **kwargs)
def to_pydict(self, *args, **kwargs):
"""
Convert the Table to a `dict` or `OrderedDict`.
Returns:
`dict`
"""
return self.table.to_pydict(*args, **kwargs)
def to_pylist(self, *args, **kwargs):
"""
Convert the Table to a list
Returns:
`list`
"""
return self.table.to_pylist(*args, **kwargs)
def to_pandas(self, *args, **kwargs):
"""
Convert to a pandas-compatible NumPy array or DataFrame, as appropriate.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
Arrow MemoryPool to use for allocations. Uses the default memory
pool is not passed.
strings_to_categorical (`bool`, defaults to `False`):
Encode string (UTF8) and binary types to `pandas.Categorical`.
categories (`list`, defaults to `empty`):
List of fields that should be returned as `pandas.Categorical`. Only
applies to table-like data structures.
zero_copy_only (`bool`, defaults to `False`):
Raise an `ArrowException` if this function call would require copying
the underlying data.
integer_object_nulls (`bool`, defaults to `False`):
Cast integers with nulls to objects.
date_as_object (`bool`, defaults to `True`):
Cast dates to objects. If `False`, convert to `datetime64[ns]` dtype.
timestamp_as_object (`bool`, defaults to `False`):
Cast non-nanosecond timestamps (`np.datetime64`) to objects. This is
useful if you have timestamps that don't fit in the normal date
range of nanosecond timestamps (1678 CE-2262 CE).
If `False`, all timestamps are converted to `datetime64[ns]` dtype.
use_threads (`bool`, defaults to `True`):
Whether to parallelize the conversion using multiple threads.
deduplicate_objects (`bool`, defaults to `False`):
Do not create multiple copies Python objects when created, to save
on memory use. Conversion will be slower.
ignore_metadata (`bool`, defaults to `False`):
If `True`, do not use the 'pandas' metadata to reconstruct the
DataFrame index, if present.
safe (`bool`, defaults to `True`):
For certain data types, a cast is needed in order to store the
data in a pandas DataFrame or Series (e.g. timestamps are always
stored as nanoseconds in pandas). This option controls whether it
is a safe cast or not.
split_blocks (`bool`, defaults to `False`):
If `True`, generate one internal "block" for each column when
creating a pandas.DataFrame from a `RecordBatch` or `Table`. While this
can temporarily reduce memory note that various pandas operations
can trigger "consolidation" which may balloon memory use.
self_destruct (`bool`, defaults to `False`):
EXPERIMENTAL: If `True`, attempt to deallocate the originating Arrow
memory while converting the Arrow object to pandas. If you use the
object after calling `to_pandas` with this option it will crash your
program.
types_mapper (`function`, defaults to `None`):
A function mapping a pyarrow DataType to a pandas `ExtensionDtype`.
This can be used to override the default pandas type for conversion
of built-in pyarrow types or in absence of `pandas_metadata` in the
Table schema. The function receives a pyarrow DataType and is
expected to return a pandas `ExtensionDtype` or `None` if the
default conversion should be used for that type. If you have
a dictionary mapping, you can pass `dict.get` as function.
Returns:
`pandas.Series` or `pandas.DataFrame`: `pandas.Series` or `pandas.DataFrame` depending on type of object
"""
return self.table.to_pandas(*args, **kwargs)
def to_string(self, *args, **kwargs):
return self.table.to_string(*args, **kwargs)
def to_reader(self, max_chunksize: Optional[int] = None):
"""
Convert the Table to a RecordBatchReader.
Note that this method is zero-copy, it merely exposes the same data under a different API.
Args:
max_chunksize (`int`, defaults to `None`)
Maximum size for RecordBatch chunks. Individual chunks may be smaller depending
on the chunk layout of individual columns.
Returns:
`pyarrow.RecordBatchReader`
"""
return self.table.to_reader(max_chunksize=max_chunksize)
def field(self, *args, **kwargs):
"""
Select a schema field by its column name or numeric index.
Args:
i (`Union[int, str]`):
The index or name of the field to retrieve.
Returns:
`pyarrow.Field`
"""
return self.table.field(*args, **kwargs)
def column(self, *args, **kwargs):
"""
Select a column by its column name, or numeric index.
Args:
i (`Union[int, str]`):
The index or name of the column to retrieve.
Returns:
`pyarrow.ChunkedArray`
"""
return self.table.column(*args, **kwargs)
def itercolumns(self, *args, **kwargs):
"""
Iterator over all columns in their numerical order.
Yields:
`pyarrow.ChunkedArray`
"""
return self.table.itercolumns(*args, **kwargs)
@property
def schema(self):
"""
Schema of the table and its columns.
Returns:
`pyarrow.Schema`
"""
return self.table.schema
@property
def columns(self):
"""
List of all columns in numerical order.
Returns:
`List[pa.ChunkedArray]`
"""
return self.table.columns
@property
def num_columns(self):
"""
Number of columns in this table.
Returns:
int
"""
return self.table.num_columns
@property
def num_rows(self):
"""
Number of rows in this table.
Due to the definition of a table, all columns have the same number of
rows.
Returns:
int
"""
return self.table.num_rows
@property
def shape(self):
"""
Dimensions of the table: (#rows, #columns).
Returns:
`(int, int)`: Number of rows and number of columns.
"""
return self.table.shape
@property
def nbytes(self):
"""
Total number of bytes consumed by the elements of the table.
"""
return self.table.nbytes
@property
def column_names(self):
"""
Names of the table's columns.
"""
return self.table.column_names
def __eq__(self, other):
return self.equals(other)
def __getitem__(self, i):
return self.table[i]
def __len__(self):
return len(self.table)
def __repr__(self):
return self.table.__repr__().replace("pyarrow.Table", self.__class__.__name__)
def __str__(self):
return self.table.__str__().replace("pyarrow.Table", self.__class__.__name__)
def slice(self, *args, **kwargs):
"""
Compute zero-copy slice of this Table.
Args:
offset (`int`, defaults to `0`):
Offset from start of table to slice.
length (`int`, defaults to `None`):
Length of slice (default is until end of table starting from
offset).
Returns:
`datasets.table.Table`
"""
raise NotImplementedError()
def filter(self, *args, **kwargs):
"""
Select records from a Table. See `pyarrow.compute.filter` for full usage.
"""
raise NotImplementedError()
def flatten(self, *args, **kwargs):
"""
Flatten this Table. Each column with a struct type is flattened
into one column per struct field. Other columns are left unchanged.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
raise NotImplementedError()
def combine_chunks(self, *args, **kwargs):
"""
Make a new table by combining the chunks this table has.
All the underlying chunks in the `ChunkedArray` of each column are
concatenated into zero or one chunk.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
raise NotImplementedError()
def cast(self, *args, **kwargs):
"""
Cast table values to another schema.
Args:
target_schema (`Schema`):
Schema to cast to, the names and order of fields must match.
safe (`bool`, defaults to `True`):
Check for overflows or other unsafe conversions.
Returns:
`datasets.table.Table`
"""
raise NotImplementedError()
def replace_schema_metadata(self, *args, **kwargs):
"""
EXPERIMENTAL: Create shallow copy of table by replacing schema
key-value metadata with the indicated new metadata (which may be None,
which deletes any existing metadata
Args:
metadata (`dict`, defaults to `None`):
Returns:
`datasets.table.Table`: shallow_copy
"""
raise NotImplementedError()
def add_column(self, *args, **kwargs):
"""
Add column to Table at position.
A new table is returned with the column added, the original table
object is left unchanged.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column added.
"""
raise NotImplementedError()
def append_column(self, *args, **kwargs):
"""
Append column at end of columns.
Args:
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column added.
"""
raise NotImplementedError()
def remove_column(self, *args, **kwargs):
"""
Create new Table with the indicated column removed.
Args:
i (`int`):
Index of column to remove.
Returns:
`datasets.table.Table`: New table without the column.
"""
raise NotImplementedError()
def set_column(self, *args, **kwargs):
"""
Replace column in Table at position.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column set.
"""
raise NotImplementedError()
def rename_columns(self, *args, **kwargs):
"""
Create new table with columns renamed to provided names.
"""
raise NotImplementedError()
def drop(self, *args, **kwargs):
"""
Drop one or more columns and return a new table.
Args:
columns (`List[str]`):
List of field names referencing existing columns.
Raises:
`KeyError` : if any of the passed columns name are not existing.
Returns:
`datasets.table.Table`: New table without the columns.
"""
raise NotImplementedError()
def select(self, *args, **kwargs):
"""
Select columns of the table.
Returns a new table with the specified columns, and metadata preserved.
Args:
columns (:obj:`Union[List[str], List[int]]`):
The column names or integer indices to select.
Returns:
`datasets.table.Table`: table with only a subset of the columns
"""
raise NotImplementedError()
class TableBlock(Table):
"""
`TableBlock` is the allowed class inside a `ConcanetationTable`.
Only `MemoryMappedTable` and `InMemoryTable` are `TableBlock`.
This is because we don't want a `ConcanetationTable` made out of other `ConcanetationTables`.
"""
pass
class InMemoryTable(TableBlock):
"""
The table is said in-memory when it is loaded into the user's RAM.
Pickling it does copy all the data using memory.
Its implementation is simple and uses the underlying pyarrow Table methods directly.
This is different from the `MemoryMapped` table, for which pickling doesn't copy all the
data in memory. For a `MemoryMapped`, unpickling instead reloads the table from the disk.
`InMemoryTable` must be used when data fit in memory, while `MemoryMapped` are reserved for
data bigger than memory or when you want the memory footprint of your application to
stay low.
"""
@classmethod
def from_file(cls, filename: str):
table = _in_memory_arrow_table_from_file(filename)
return cls(table)
@classmethod
def from_buffer(cls, buffer: pa.Buffer):
table = _in_memory_arrow_table_from_buffer(buffer)
return cls(table)
@classmethod
def from_pandas(cls, *args, **kwargs):
"""
Convert pandas.DataFrame to an Arrow Table.
The column types in the resulting Arrow Table are inferred from the
dtypes of the pandas.Series in the DataFrame. In the case of non-object
Series, the NumPy dtype is translated to its Arrow equivalent. In the
case of `object`, we need to guess the datatype by looking at the
Python objects in this Series.
Be aware that Series of the `object` dtype don't carry enough
information to always lead to a meaningful Arrow type. In the case that
we cannot infer a type, e.g. because the DataFrame is of length 0 or
the Series only contains `None/nan` objects, the type is set to
null. This behavior can be avoided by constructing an explicit schema
and passing it to this function.
Args:
df (`pandas.DataFrame`):
schema (`pyarrow.Schema`, *optional*):
The expected schema of the Arrow Table. This can be used to
indicate the type of columns if we cannot infer it automatically.
If passed, the output will have exactly this schema. Columns
specified in the schema that are not found in the DataFrame columns
or its index will raise an error. Additional columns or index
levels in the DataFrame which are not specified in the schema will
be ignored.
preserve_index (`bool`, *optional*):
Whether to store the index as an additional column in the resulting
`Table`. The default of None will store the index as a column,
except for RangeIndex which is stored as metadata only. Use
`preserve_index=True` to force it to be stored as a column.
nthreads (`int`, defaults to `None` (may use up to system CPU count threads))
If greater than 1, convert columns to Arrow in parallel using
indicated number of threads.
columns (`List[str]`, *optional*):
List of column to be converted. If `None`, use all columns.
safe (`bool`, defaults to `True`):
Check for overflows or other unsafe conversions,
Returns:
`datasets.table.Table`:
Examples:
```python
>>> import pandas as pd
>>> import pyarrow as pa
>>> df = pd.DataFrame({
... 'int': [1, 2],
... 'str': ['a', 'b']
... })
>>> pa.Table.from_pandas(df)
<pyarrow.lib.Table object at 0x7f05d1fb1b40>
```
"""
return cls(pa.Table.from_pandas(*args, **kwargs))
@classmethod
def from_arrays(cls, *args, **kwargs):
"""
Construct a Table from Arrow arrays.
Args:
arrays (`List[Union[pyarrow.Array, pyarrow.ChunkedArray]]`):
Equal-length arrays that should form the table.
names (`List[str]`, *optional*):
Names for the table columns. If not passed, schema must be passed.
schema (`Schema`, defaults to `None`):
Schema for the created table. If not passed, names must be passed.
metadata (`Union[dict, Mapping]`, defaults to `None`):
Optional metadata for the schema (if inferred).
Returns:
`datasets.table.Table`
"""
return cls(pa.Table.from_arrays(*args, **kwargs))
@classmethod
def from_pydict(cls, *args, **kwargs):
"""
Construct a Table from Arrow arrays or columns.
Args:
mapping (`Union[dict, Mapping]`):
A mapping of strings to Arrays or Python lists.
schema (`Schema`, defaults to `None`):
If not passed, will be inferred from the Mapping values
metadata (`Union[dict, Mapping]`, defaults to `None`):
Optional metadata for the schema (if inferred).
Returns:
`datasets.table.Table`
"""
return cls(pa.Table.from_pydict(*args, **kwargs))
@classmethod
def from_pylist(cls, mapping, *args, **kwargs):
"""
Construct a Table from list of rows / dictionaries.
Args:
mapping (`List[dict]`):
A mapping of strings to row values.
schema (`Schema`, defaults to `None`):
If not passed, will be inferred from the Mapping values
metadata (`Union[dict, Mapping]`, defaults to `None`):
Optional metadata for the schema (if inferred).
Returns:
`datasets.table.Table`
"""
return cls(pa.Table.from_pylist(mapping, *args, **kwargs))
@classmethod
def from_batches(cls, *args, **kwargs):
"""
Construct a Table from a sequence or iterator of Arrow `RecordBatches`.
Args:
batches (`Union[Sequence[pyarrow.RecordBatch], Iterator[pyarrow.RecordBatch]]`):
Sequence of `RecordBatch` to be converted, all schemas must be equal.
schema (`Schema`, defaults to `None`):
If not passed, will be inferred from the first `RecordBatch`.
Returns:
`datasets.table.Table`:
"""
return cls(pa.Table.from_batches(*args, **kwargs))
def slice(self, offset=0, length=None):
"""
Compute zero-copy slice of this Table.
Args:
offset (`int`, defaults to `0`):
Offset from start of table to slice.
length (`int`, defaults to `None`):
Length of slice (default is until end of table starting from
offset).
Returns:
`datasets.table.Table`
"""
# Use fast slicing here
return InMemoryTable(self.fast_slice(offset=offset, length=length))
def filter(self, *args, **kwargs):
"""
Select records from a Table. See `pyarrow.compute.filter` for full usage.
"""
return InMemoryTable(self.table.filter(*args, **kwargs))
def flatten(self, *args, **kwargs):
"""
Flatten this Table. Each column with a struct type is flattened
into one column per struct field. Other columns are left unchanged.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
return InMemoryTable(table_flatten(self.table, *args, **kwargs))
def combine_chunks(self, *args, **kwargs):
"""
Make a new table by combining the chunks this table has.
All the underlying chunks in the `ChunkedArray` of each column are
concatenated into zero or one chunk.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
return InMemoryTable(self.table.combine_chunks(*args, **kwargs))
def cast(self, *args, **kwargs):
"""
Cast table values to another schema.
Args:
target_schema (`Schema`):
Schema to cast to, the names and order of fields must match.
safe (`bool`, defaults to `True`):
Check for overflows or other unsafe conversions.
Returns:
`datasets.table.Table`
"""
return InMemoryTable(table_cast(self.table, *args, **kwargs))
def replace_schema_metadata(self, *args, **kwargs):
"""
EXPERIMENTAL: Create shallow copy of table by replacing schema
key-value metadata with the indicated new metadata (which may be `None`,
which deletes any existing metadata).
Args:
metadata (`dict`, defaults to `None`):
Returns:
`datasets.table.Table`: shallow_copy
"""
return InMemoryTable(self.table.replace_schema_metadata(*args, **kwargs))
def add_column(self, *args, **kwargs):
"""
Add column to Table at position.
A new table is returned with the column added, the original table
object is left unchanged.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column added.
"""
return InMemoryTable(self.table.add_column(*args, **kwargs))
def append_column(self, *args, **kwargs):
"""
Append column at end of columns.
Args:
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column added.
"""
return InMemoryTable(self.table.append_column(*args, **kwargs))
def remove_column(self, *args, **kwargs):
"""
Create new Table with the indicated column removed.
Args:
i (`int`):
Index of column to remove.
Returns:
`datasets.table.Table`:
New table without the column.
"""
return InMemoryTable(self.table.remove_column(*args, **kwargs))
def set_column(self, *args, **kwargs):
"""
Replace column in Table at position.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column set.
"""
return InMemoryTable(self.table.set_column(*args, **kwargs))
def rename_columns(self, *args, **kwargs):
"""
Create new table with columns renamed to provided names.
"""
return InMemoryTable(self.table.rename_columns(*args, **kwargs))
def drop(self, *args, **kwargs):
"""
Drop one or more columns and return a new table.
Args:
columns (`List[str]`):
List of field names referencing existing columns.
Raises:
`KeyError` : if any of the passed columns name are not existing.
Returns:
`datasets.table.Table`:
New table without the columns.
"""
return InMemoryTable(self.table.drop(*args, **kwargs))
def select(self, *args, **kwargs):
"""
Select columns of the table.
Returns a new table with the specified columns, and metadata preserved.
Args:
columns (:obj:`Union[List[str], List[int]]`):
The column names or integer indices to select.
Returns:
:class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
"""
return InMemoryTable(self.table.select(*args, **kwargs))
# The MemoryMappedTable needs replays to properly reload tables from the disk
Replay = tuple[str, tuple, dict]
class MemoryMappedTable(TableBlock):
"""
The table is said memory mapped when it doesn't use the user's RAM but loads the data
from the disk instead.
Pickling it doesn't copy the data into memory.
Instead, only the path to the memory mapped arrow file is pickled, as well as the list
of transforms to "replay" when reloading the table from the disk.
Its implementation requires to store an history of all the transforms that were applied
to the underlying pyarrow Table, so that they can be "replayed" when reloading the Table
from the disk.
This is different from the `InMemoryTable` table, for which pickling does copy all the
data in memory.
`InMemoryTable` must be used when data fit in memory, while `MemoryMapped` are reserved for
data bigger than memory or when you want the memory footprint of your application to
stay low.
"""
def __init__(self, table: pa.Table, path: str, replays: Optional[list[Replay]] = None):
super().__init__(table)
self.path = os.path.abspath(path)
self.replays: list[Replay] = replays if replays is not None else []
@classmethod
def from_file(cls, filename: str, replays=None):
table = _memory_mapped_arrow_table_from_file(filename)
table = cls._apply_replays(table, replays)
return cls(table, filename, replays)
def __getstate__(self):
return {"path": self.path, "replays": self.replays}
def __setstate__(self, state):
path = state["path"]
replays = state["replays"]
table = _memory_mapped_arrow_table_from_file(path)
table = self._apply_replays(table, replays)
MemoryMappedTable.__init__(self, table, path=path, replays=replays)
@staticmethod
def _apply_replays(table: pa.Table, replays: Optional[list[Replay]] = None) -> pa.Table:
if replays is not None:
for name, args, kwargs in replays:
if name == "cast":
table = table_cast(table, *args, **kwargs)
elif name == "flatten":
table = table_flatten(table, *args, **kwargs)
else:
table = getattr(table, name)(*args, **kwargs)
return table
def _append_replay(self, replay: Replay) -> list[Replay]:
replays = copy.deepcopy(self.replays)
replays.append(replay)
return replays
def slice(self, offset=0, length=None):
"""
Compute zero-copy slice of this Table.
Args:
offset (`int`, defaults to `0`):
Offset from start of table to slice.
length (`int`, defaults to `None`):
Length of slice (default is until end of table starting from
offset).
Returns:
`datasets.table.Table`
"""
replay = ("slice", (offset, length), {})
replays = self._append_replay(replay)
# Use fast slicing here
return MemoryMappedTable(self.fast_slice(offset=offset, length=length), self.path, replays)
def filter(self, *args, **kwargs):
"""
Select records from a Table. See `pyarrow.compute.filter` for full usage.
"""
replay = ("filter", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.filter(*args, **kwargs), self.path, replays)
def flatten(self, *args, **kwargs):
"""
Flatten this Table. Each column with a struct type is flattened
into one column per struct field. Other columns are left unchanged.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
replay = ("flatten", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(table_flatten(self.table, *args, **kwargs), self.path, replays)
def combine_chunks(self, *args, **kwargs):
"""
Make a new table by combining the chunks this table has.
All the underlying chunks in the ChunkedArray of each column are
concatenated into zero or one chunk.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
replay = ("combine_chunks", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.combine_chunks(*args, **kwargs), self.path, replays)
def cast(self, *args, **kwargs):
"""
Cast table values to another schema
Args:
target_schema (`Schema`):
Schema to cast to, the names and order of fields must match.
safe (`bool`, defaults to `True`):
Check for overflows or other unsafe conversions.
Returns:
`datasets.table.Table`
"""
replay = ("cast", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(table_cast(self.table, *args, **kwargs), self.path, replays)
def replace_schema_metadata(self, *args, **kwargs):
"""
EXPERIMENTAL: Create shallow copy of table by replacing schema
key-value metadata with the indicated new metadata (which may be None,
which deletes any existing metadata.
Args:
metadata (`dict`, defaults to `None`):
Returns:
`datasets.table.Table`: shallow_copy
"""
replay = ("replace_schema_metadata", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.replace_schema_metadata(*args, **kwargs), self.path, replays)
def add_column(self, *args, **kwargs):
"""
Add column to Table at position.
A new table is returned with the column added, the original table
object is left unchanged.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column added.
"""
replay = ("add_column", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.add_column(*args, **kwargs), self.path, replays)
def append_column(self, *args, **kwargs):
"""
Append column at end of columns.
Args:
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column added.
"""
replay = ("append_column", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.append_column(*args, **kwargs), self.path, replays)
def remove_column(self, *args, **kwargs):
"""
Create new Table with the indicated column removed.
Args:
i (`int`):
Index of column to remove.
Returns:
`datasets.table.Table`:
New table without the column.
"""
replay = ("remove_column", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.remove_column(*args, **kwargs), self.path, replays)
def set_column(self, *args, **kwargs):
"""
Replace column in Table at position.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column set.
"""
replay = ("set_column", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.set_column(*args, **kwargs), self.path, replays)
def rename_columns(self, *args, **kwargs):
"""
Create new table with columns renamed to provided names.
"""
replay = ("rename_columns", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.rename_columns(*args, **kwargs), self.path, replays)
def drop(self, *args, **kwargs):
"""
Drop one or more columns and return a new table.
Args:
columns (`List[str]`):
List of field names referencing existing columns.
Raises:
`KeyError` : if any of the passed columns name are not existing.
Returns:
`datasets.table.Table`:
New table without the columns.
"""
replay = ("drop", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.drop(*args, **kwargs), self.path, replays)
def select(self, *args, **kwargs):
"""
Select columns of the table.
Returns a new table with the specified columns, and metadata preserved.
Args:
columns (:obj:`Union[List[str], List[int]]`):
The column names or integer indices to select.
Returns:
:class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
"""
replay = ("select", copy.deepcopy(args), copy.deepcopy(kwargs))
replays = self._append_replay(replay)
return MemoryMappedTable(self.table.select(*args, **kwargs), self.path, replays)
# A ConcatenationTable is the concatenation of several tables.
# The ``blocks`` attributes stores a list of list of blocks.
# The first axis concatenates the tables along the axis 0 (it appends rows),
# while the second axis concatenates tables along the axis 1 (it appends columns).
TableBlockContainer = TypeVar("TableBlockContainer", TableBlock, list[TableBlock], list[list[TableBlock]])
class ConcatenationTable(Table):
"""
The table comes from the concatenation of several tables called blocks.
It enables concatenation on both axis 0 (append rows) and axis 1 (append columns).
The underlying tables are called "blocks" and can be either `InMemoryTable`
or `MemoryMappedTable` objects.
This allows to combine tables that come from memory or that are memory mapped.
When a `ConcatenationTable` is pickled, then each block is pickled:
- the `InMemoryTable` objects are pickled by copying all the data in memory.
- the MemoryMappedTable objects are pickled without copying the data into memory.
Instead, only the path to the memory mapped arrow file is pickled, as well as the list
of transforms to "replays" when reloading the table from the disk.
Its implementation requires to store each block separately.
The `blocks` attributes stores a list of list of blocks.
The first axis concatenates the tables along the axis 0 (it appends rows),
while the second axis concatenates tables along the axis 1 (it appends columns).
If some columns are missing when concatenating on axis 0, they are filled with null values.
This is done using `pyarrow.concat_tables(tables, promote=True)`.
You can access the fully combined table by accessing the `ConcatenationTable.table` attribute,
and the blocks by accessing the `ConcatenationTable.blocks` attribute.
"""
def __init__(self, table: pa.Table, blocks: list[list[TableBlock]]):
super().__init__(table)
self.blocks = blocks
# Check that all the blocks have the right type.
# Only InMemoryTable and MemoryMappedTable are allowed.
for subtables in blocks:
for subtable in subtables:
if not isinstance(subtable, TableBlock):
raise TypeError(
"The blocks of a ConcatenationTable must be InMemoryTable or MemoryMappedTable objects"
f", but got {_short_str(subtable)}."
)
def __getstate__(self):
return {"blocks": self.blocks, "schema": self.table.schema}
def __setstate__(self, state):
blocks = state["blocks"]
schema = state["schema"]
table = self._concat_blocks_horizontally_and_vertically(blocks)
if schema is not None and table.schema != schema:
# We fix the columns by concatenating with an empty table with the right columns
empty_table = pa.Table.from_batches([], schema=schema)
# We set promote_options="default" to fill missing columns with null values
table = pa.concat_tables([table, empty_table], promote_options="default")
ConcatenationTable.__init__(self, table, blocks=blocks)
@staticmethod
def _concat_blocks(blocks: list[Union[TableBlock, pa.Table]], axis: int = 0) -> pa.Table:
pa_tables = [table.table if hasattr(table, "table") else table for table in blocks]
if axis == 0:
# We set promote_options="default" to fill missing columns with null values
return pa.concat_tables(pa_tables, promote_options="default")
elif axis == 1:
for i, table in enumerate(pa_tables):
if i == 0:
pa_table = table
else:
for name, col in zip(table.column_names, table.columns):
pa_table = pa_table.append_column(name, col)
return pa_table
else:
raise ValueError("'axis' must be either 0 or 1")
@classmethod
def _concat_blocks_horizontally_and_vertically(cls, blocks: list[list[TableBlock]]) -> pa.Table:
pa_tables_to_concat_vertically = []
for i, tables in enumerate(blocks):
if not tables:
continue
pa_table_horizontally_concatenated = cls._concat_blocks(tables, axis=1)
pa_tables_to_concat_vertically.append(pa_table_horizontally_concatenated)
return cls._concat_blocks(pa_tables_to_concat_vertically, axis=0)
@classmethod
def _merge_blocks(cls, blocks: TableBlockContainer, axis: Optional[int] = None) -> TableBlockContainer:
if axis is not None:
merged_blocks = []
for is_in_memory, block_group in groupby(blocks, key=lambda x: isinstance(x, InMemoryTable)):
if is_in_memory:
block_group = [InMemoryTable(cls._concat_blocks(list(block_group), axis=axis))]
merged_blocks += list(block_group)
else: # both
merged_blocks = [cls._merge_blocks(row_block, axis=1) for row_block in blocks]
if all(len(row_block) == 1 for row_block in merged_blocks):
merged_blocks = cls._merge_blocks(
[block for row_block in merged_blocks for block in row_block], axis=0
)
return merged_blocks
@classmethod
def _consolidate_blocks(cls, blocks: TableBlockContainer) -> TableBlockContainer:
if isinstance(blocks, TableBlock):
return blocks
elif isinstance(blocks[0], TableBlock):
return cls._merge_blocks(blocks, axis=0)
else:
return cls._merge_blocks(blocks)
@classmethod
def from_blocks(cls, blocks: TableBlockContainer) -> "ConcatenationTable":
blocks = cls._consolidate_blocks(blocks)
if isinstance(blocks, TableBlock):
table = blocks
return cls(table.table, [[table]])
elif isinstance(blocks[0], TableBlock):
table = cls._concat_blocks(blocks, axis=0)
blocks = [[t] for t in blocks]
return cls(table, blocks)
else:
table = cls._concat_blocks_horizontally_and_vertically(blocks)
return cls(table, blocks)
@classmethod
def from_tables(cls, tables: list[Union[pa.Table, Table]], axis: int = 0) -> "ConcatenationTable":
"""Create `ConcatenationTable` from list of tables.
Args:
tables (list of `Table` or list of `pyarrow.Table`):
List of tables.
axis (`{0, 1}`, defaults to `0`, meaning over rows):
Axis to concatenate over, where `0` means over rows (vertically) and `1` means over columns
(horizontally).
<Added version="1.6.0"/>
"""
def to_blocks(table: Union[pa.Table, Table]) -> list[list[TableBlock]]:
if isinstance(table, pa.Table):
return [[InMemoryTable(table)]]
elif isinstance(table, ConcatenationTable):
return copy.deepcopy(table.blocks)
else:
return [[table]]
def _slice_row_block(row_block: list[TableBlock], length: int) -> tuple[list[TableBlock], list[TableBlock]]:
sliced = [table.slice(0, length) for table in row_block]
remainder = [table.slice(length, len(row_block[0]) - length) for table in row_block]
return sliced, remainder
def _split_both_like(
result: list[list[TableBlock]], blocks: list[list[TableBlock]]
) -> tuple[list[list[TableBlock]], list[list[TableBlock]]]:
"""
Make sure each row_block contain the same num_rows to be able to concatenate them on axis=1.
To do so, we modify both blocks sets to have the same row_blocks boundaries.
For example, if `result` has 2 row_blocks of 3 rows and `blocks` has 3 row_blocks of 2 rows,
we modify both to have 4 row_blocks of size 2, 1, 1 and 2:
[ x x x | x x x ]
+ [ y y | y y | y y ]
-----------------------------
= [ x x | x | x | x x ]
[ y y | y | y | y y ]
"""
result, blocks = list(result), list(blocks)
new_result, new_blocks = [], []
while result and blocks:
# we slice the longest row block to save two row blocks of same length
# and we replace the long row block by its remainder if necessary
if len(result[0][0]) > len(blocks[0][0]):
new_blocks.append(blocks[0])
sliced, result[0] = _slice_row_block(result[0], len(blocks.pop(0)[0]))
new_result.append(sliced)
elif len(result[0][0]) < len(blocks[0][0]):
new_result.append(result[0])
sliced, blocks[0] = _slice_row_block(blocks[0], len(result.pop(0)[0]))
new_blocks.append(sliced)
else:
new_result.append(result.pop(0))
new_blocks.append(blocks.pop(0))
if result or blocks:
raise ValueError("Failed to concatenate on axis=1 because tables don't have the same number of rows")
return new_result, new_blocks
def _extend_blocks(
result: list[list[TableBlock]], blocks: list[list[TableBlock]], axis: int = 0
) -> list[list[TableBlock]]:
if axis == 0:
result.extend(blocks)
elif axis == 1:
# We make sure each row_block have the same num_rows
result, blocks = _split_both_like(result, blocks)
for i, row_block in enumerate(blocks):
result[i].extend(row_block)
return result
blocks = to_blocks(tables[0])
for table in tables[1:]:
table_blocks = to_blocks(table)
blocks = _extend_blocks(blocks, table_blocks, axis=axis)
return cls.from_blocks(blocks)
@property
def _slices(self):
offset = 0
for tables in self.blocks:
length = len(tables[0])
yield (offset, length)
offset += length
def slice(self, offset=0, length=None):
"""
Compute zero-copy slice of this Table.
Args:
offset (`int`, defaults to `0`):
Offset from start of table to slice.
length (`int`, defaults to `None`):
Length of slice (default is until end of table starting from
offset).
Returns:
`datasets.table.Table`
"""
table = self.table.slice(offset, length=length)
length = length if length is not None else self.num_rows - offset
blocks = []
for tables in self.blocks:
n_rows = len(tables[0])
if length == 0:
break
elif n_rows <= offset:
offset = offset - n_rows
elif n_rows <= offset + length:
blocks.append([t.slice(offset) for t in tables])
length, offset = length + offset - n_rows, 0
else:
blocks.append([t.slice(offset, length) for t in tables])
length, offset = 0, 0
return ConcatenationTable(table, blocks)
def filter(self, mask, *args, **kwargs):
"""
Select records from a Table. See `pyarrow.compute.filter` for full usage.
"""
table = self.table.filter(mask, *args, **kwargs)
blocks = []
for (offset, length), tables in zip(self._slices, self.blocks):
submask = mask.slice(offset, length)
blocks.append([t.filter(submask, *args, **kwargs) for t in tables])
return ConcatenationTable(table, blocks)
def flatten(self, *args, **kwargs):
"""
Flatten this Table. Each column with a struct type is flattened
into one column per struct field. Other columns are left unchanged.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
table = table_flatten(self.table, *args, **kwargs)
blocks = []
for tables in self.blocks:
blocks.append([t.flatten(*args, **kwargs) for t in tables])
return ConcatenationTable(table, blocks)
def combine_chunks(self, *args, **kwargs):
"""
Make a new table by combining the chunks this table has.
All the underlying chunks in the `ChunkedArray` of each column are
concatenated into zero or one chunk.
Args:
memory_pool (`MemoryPool`, defaults to `None`):
For memory allocations, if required, otherwise use default pool.
Returns:
`datasets.table.Table`
"""
table = self.table.combine_chunks(*args, **kwargs)
blocks = []
for tables in self.blocks:
blocks.append([t.combine_chunks(*args, **kwargs) for t in tables])
return ConcatenationTable(table, blocks)
def cast(self, target_schema, *args, **kwargs):
"""
Cast table values to another schema.
Args:
target_schema (`Schema`):
Schema to cast to, the names and order of fields must match.
safe (`bool`, defaults to `True`):
Check for overflows or other unsafe conversions.
Returns:
`datasets.table.Table`
"""
from .features import Features
table = table_cast(self.table, target_schema, *args, **kwargs)
target_features = Features.from_arrow_schema(target_schema)
blocks = []
for subtables in self.blocks:
new_tables = []
fields = list(target_schema)
for subtable in subtables:
subfields = []
for name in subtable.column_names:
subfields.append(fields.pop(next(i for i, field in enumerate(fields) if field.name == name)))
subfeatures = Features({subfield.name: target_features[subfield.name] for subfield in subfields})
subschema = subfeatures.arrow_schema
new_tables.append(subtable.cast(subschema, *args, **kwargs))
blocks.append(new_tables)
return ConcatenationTable(table, blocks)
def replace_schema_metadata(self, *args, **kwargs):
"""
EXPERIMENTAL: Create shallow copy of table by replacing schema
key-value metadata with the indicated new metadata (which may be `None`,
which deletes any existing metadata).
Args:
metadata (`dict`, defaults to `None`):
Returns:
`datasets.table.Table`: shallow_copy
"""
table = self.table.replace_schema_metadata(*args, **kwargs)
blocks = []
for tables in self.blocks:
blocks.append([t.replace_schema_metadata(*args, **kwargs) for t in tables])
return ConcatenationTable(table, self.blocks)
def add_column(self, *args, **kwargs):
"""
Add column to Table at position.
A new table is returned with the column added, the original table
object is left unchanged.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`: New table with the passed column added.
"""
raise NotImplementedError()
def append_column(self, *args, **kwargs):
"""
Append column at end of columns.
Args:
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column added.
"""
raise NotImplementedError()
def remove_column(self, i, *args, **kwargs):
"""
Create new Table with the indicated column removed.
Args:
i (`int`):
Index of column to remove.
Returns:
`datasets.table.Table`:
New table without the column.
"""
table = self.table.remove_column(i, *args, **kwargs)
name = self.table.column_names[i]
blocks = []
for tables in self.blocks:
blocks.append(
[
t.remove_column(t.column_names.index(name), *args, **kwargs) if name in t.column_names else t
for t in tables
]
)
return ConcatenationTable(table, blocks)
def set_column(self, *args, **kwargs):
"""
Replace column in Table at position.
Args:
i (`int`):
Index to place the column at.
field_ (`Union[str, pyarrow.Field]`):
If a string is passed then the type is deduced from the column
data.
column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
Column data.
Returns:
`datasets.table.Table`:
New table with the passed column set.
"""
raise NotImplementedError()
def rename_columns(self, names, *args, **kwargs):
"""
Create new table with columns renamed to provided names.
"""
table = self.table.rename_columns(names, *args, **kwargs)
names = dict(zip(self.table.column_names, names))
blocks = []
for tables in self.blocks:
blocks.append(
[t.rename_columns([names[name] for name in t.column_names], *args, **kwargs) for t in tables]
)
return ConcatenationTable(table, blocks)
def drop(self, columns, *args, **kwargs):
"""
Drop one or more columns and return a new table.
Args:
columns (`List[str]`):
List of field names referencing existing columns.
Raises:
`KeyError` : if any of the passed columns name are not existing.
Returns:
`datasets.table.Table`:
New table without the columns.
"""
table = self.table.drop(columns, *args, **kwargs)
blocks = []
for tables in self.blocks:
blocks.append([t.drop([c for c in columns if c in t.column_names], *args, **kwargs) for t in tables])
return ConcatenationTable(table, blocks)
def select(self, columns, *args, **kwargs):
"""
Select columns of the table.
Returns a new table with the specified columns, and metadata preserved.
Args:
columns (:obj:`Union[List[str], List[int]]`):
The column names or integer indices to select.
Returns:
:class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
"""
table = self.table.select(columns, *args, **kwargs)
blocks = []
for tables in self.blocks:
blocks.append([t.select([c for c in columns if c in t.column_names], *args, **kwargs) for t in tables])
return ConcatenationTable(table, blocks)
def concat_tables(tables: list[Table], axis: int = 0) -> Table:
"""
Concatenate tables.
Args:
tables (list of `Table`):
List of tables to be concatenated.
axis (`{0, 1}`, defaults to `0`, meaning over rows):
Axis to concatenate over, where `0` means over rows (vertically) and `1` means over columns
(horizontally).
<Added version="1.6.0"/>
Returns:
`datasets.table.Table`:
If the number of input tables is > 1, then the returned table is a `datasets.table.ConcatenationTable`.
Otherwise if there's only one table, it is returned as is.
"""
tables = list(tables)
if len(tables) == 1:
return tables[0]
return ConcatenationTable.from_tables(tables, axis=axis)
def list_table_cache_files(table: Table) -> list[str]:
"""
Get the cache files that are loaded by the table.
Cache file are used when parts of the table come from the disk via memory mapping.
Returns:
`List[str]`:
A list of paths to the cache files loaded by the table.
"""
if isinstance(table, ConcatenationTable):
cache_files = []
for subtables in table.blocks:
for subtable in subtables:
cache_files += list_table_cache_files(subtable)
return cache_files
elif isinstance(table, MemoryMappedTable):
return [table.path]
else:
return []
def _wrap_for_chunked_arrays(func):
"""Apply the function on each chunk of a `pyarrow.ChunkedArray`, or on the array directly"""
def wrapper(array, *args, **kwargs):
if isinstance(array, pa.ChunkedArray):
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
else:
return func(array, *args, **kwargs)
return wrapper
def _are_list_values_of_length(array: pa.ListArray, length: int) -> bool:
"""Check if all the sub-lists of a `pa.ListArray` have the specified length."""
return pc.all(pc.equal(array.value_lengths(), length)).as_py() or array.null_count == len(array)
def _combine_list_array_offsets_with_mask(array: pa.ListArray) -> pa.Array:
"""Add the null bitmap to the offsets of a `pa.ListArray`."""
offsets = array.offsets
if array.null_count > 0:
offsets = pa.concat_arrays(
[
pc.replace_with_mask(offsets[:-1], array.is_null(), pa.nulls(len(array), pa.int32())),
offsets[-1:],
]
)
return offsets
def _storage_type(type: pa.DataType) -> pa.DataType:
"""Convert a (possibly nested) `pa.ExtensionType` to its storage type."""
if isinstance(type, pa.ExtensionType):
return _storage_type(type.storage_type)
elif isinstance(type, pa.StructType):
return pa.struct([pa.field(field.name, _storage_type(field.type)) for field in type])
elif isinstance(type, pa.ListType):
return pa.list_(_storage_type(type.value_type))
elif isinstance(type, pa.FixedSizeListType):
return pa.list_(_storage_type(type.value_type), type.list_size)
return type
def _short_str(value: Any) -> str:
out = str(value)
if len(out) > 3000:
out = out[:1500] + "\n...\n" + out[-1500:]
return out
@_wrap_for_chunked_arrays
def array_cast(
array: pa.Array, pa_type: pa.DataType, allow_primitive_to_str: bool = True, allow_decimal_to_str: bool = True
) -> Union[pa.Array, pa.FixedSizeListArray, pa.ListArray, pa.StructArray, pa.ExtensionArray]:
"""Improved version of `pa.Array.cast`
It supports casting `pa.StructArray` objects to re-order the fields.
It also let you control certain aspects of the casting, e.g. whether
to disable casting primitives (`booleans`, `floats` or `ints`) or
disable casting decimals to strings.
Args:
array (`pa.Array`):
PyArrow array to cast
pa_type (`pa.DataType`):
Target PyArrow type
allow_primitive_to_str (`bool`, defaults to `True`):
Whether to allow casting primitives to strings.
Defaults to `True`.
allow_decimal_to_str (`bool`, defaults to `True`):
Whether to allow casting decimals to strings.
Defaults to `True`.
Raises:
`pa.ArrowInvalidError`: if the arrow data casting fails
`TypeError`: if the target type is not supported according, e.g.
- if a field is missing
- if casting from primitives to strings and `allow_primitive_to_str` is `False`
- if casting from decimals to strings and `allow_decimal_to_str` is `False`
Returns:
`List[pyarrow.Array]`: the casted array
"""
_c = partial(array_cast, allow_primitive_to_str=allow_primitive_to_str, allow_decimal_to_str=allow_decimal_to_str)
if isinstance(array, pa.ExtensionArray):
array = array.storage
if isinstance(pa_type, pa.ExtensionType):
return pa_type.wrap_array(_c(array, pa_type.storage_type))
elif array.type == pa_type:
return array
elif pa.types.is_struct(array.type):
if pa.types.is_struct(pa_type) and ({field.name for field in pa_type} == {field.name for field in array.type}):
if array.type.num_fields == 0:
return array
arrays = [_c(array.field(field.name), field.type) for field in pa_type]
return pa.StructArray.from_arrays(arrays, fields=list(pa_type), mask=array.is_null())
elif pa.types.is_list(array.type) or pa.types.is_large_list(array.type):
if pa.types.is_fixed_size_list(pa_type):
if _are_list_values_of_length(array, pa_type.list_size):
if array.null_count > 0:
# Ensure each null value in the array translates to [null] * pa_type.list_size in the array's values array
array_type = array.type
storage_type = _storage_type(array_type)
if array_type != storage_type:
# Temporarily convert to the storage type to support extension types in the slice operation
array = _c(array, storage_type)
array = pc.list_slice(array, 0, pa_type.list_size, return_fixed_size_list=True)
array = _c(array, array_type)
else:
array = pc.list_slice(array, 0, pa_type.list_size, return_fixed_size_list=True)
array_values = array.values
return pa.FixedSizeListArray.from_arrays(
_c(array_values, pa_type.value_type), pa_type.list_size, mask=array.is_null()
)
else:
array_values = array.values[
array.offset * pa_type.list_size : (array.offset + len(array)) * pa_type.list_size
]
return pa.FixedSizeListArray.from_arrays(_c(array_values, pa_type.value_type), pa_type.list_size)
elif pa.types.is_list(pa_type):
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.ListArray.from_arrays(array_offsets, _c(array.values, pa_type.value_type))
elif pa.types.is_large_list(pa_type):
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.LargeListArray.from_arrays(array_offsets, _c(array.values, pa_type.value_type))
elif pa.types.is_fixed_size_list(array.type):
if pa.types.is_fixed_size_list(pa_type):
if pa_type.list_size == array.type.list_size:
array_values = array.values[
array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
]
return pa.FixedSizeListArray.from_arrays(
_c(array_values, pa_type.value_type), pa_type.list_size, mask=array.is_null()
)
elif pa.types.is_list(pa_type):
array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
return pa.ListArray.from_arrays(array_offsets, _c(array.values, pa_type.value_type), mask=array.is_null())
elif pa.types.is_large_list(pa_type):
array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
return pa.LargeListArray.from_arrays(
array_offsets, _c(array.values, pa_type.value_type), mask=array.is_null()
)
else:
if pa.types.is_string(pa_type):
if not allow_primitive_to_str and pa.types.is_primitive(array.type):
raise TypeError(
f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)} "
f"since allow_primitive_to_str is set to {allow_primitive_to_str} "
)
if not allow_decimal_to_str and pa.types.is_decimal(array.type):
raise TypeError(
f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)} "
f"and allow_decimal_to_str is set to {allow_decimal_to_str}"
)
if pa.types.is_null(pa_type) and not pa.types.is_null(array.type):
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
return array.cast(pa_type)
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
@_wrap_for_chunked_arrays
def cast_array_to_feature(
array: pa.Array, feature: "FeatureType", allow_primitive_to_str: bool = True, allow_decimal_to_str: bool = True
) -> pa.Array:
"""Cast an array to the arrow type that corresponds to the requested feature type.
For custom features like [`Audio`] or [`Image`], it takes into account the "cast_storage" methods
they defined to enable casting from other arrow types.
Args:
array (`pa.Array`):
The PyArrow array to cast.
feature (`datasets.features.FeatureType`):
The target feature type.
allow_primitive_to_str (`bool`, defaults to `True`):
Whether to allow casting primitives to strings.
Defaults to `True`.
allow_decimal_to_str (`bool`, defaults to `True`):
Whether to allow casting decimals to strings.
Defaults to `True`.
Raises:
`pa.ArrowInvalidError`: if the arrow data casting fails
`TypeError`: if the target type is not supported according, e.g.
- if a field is missing
- if casting from primitives and `allow_primitive_to_str` is `False`
- if casting from decimals and `allow_decimal_to_str` is `False`
Returns:
array (`pyarrow.Array`): the casted array
"""
from .features.features import LargeList, Sequence, get_nested_type
_c = partial(
cast_array_to_feature,
allow_primitive_to_str=allow_primitive_to_str,
allow_decimal_to_str=allow_decimal_to_str,
)
if isinstance(array, pa.ExtensionArray):
array = array.storage
if hasattr(feature, "cast_storage"):
return feature.cast_storage(array)
elif pa.types.is_struct(array.type):
# feature must be a dict or Sequence(subfeatures_dict)
if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
sequence_kwargs = vars(feature).copy()
feature = sequence_kwargs.pop("feature")
feature = {name: Sequence(subfeature, **sequence_kwargs) for name, subfeature in feature.items()}
if isinstance(feature, dict) and (array_fields := {field.name for field in array.type}) <= set(feature):
null_array = pa.array([None] * len(array))
arrays = [
_c(array.field(name) if name in array_fields else null_array, subfeature)
for name, subfeature in feature.items()
]
return pa.StructArray.from_arrays(arrays, names=list(feature), mask=array.is_null())
elif pa.types.is_list(array.type) or pa.types.is_large_list(array.type):
# feature must be either [subfeature] or LargeList(subfeature) or Sequence(subfeature)
if isinstance(feature, list):
casted_array_values = _c(array.values, feature[0])
if pa.types.is_list(array.type) and casted_array_values.type == array.values.type:
# Both array and feature have equal list type and values (within the list) type
return array
else:
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.ListArray.from_arrays(array_offsets, casted_array_values)
elif isinstance(feature, LargeList):
casted_array_values = _c(array.values, feature.feature)
if pa.types.is_large_list(array.type) and casted_array_values.type == array.values.type:
# Both array and feature have equal large_list type and values (within the list) type
return array
else:
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.LargeListArray.from_arrays(array_offsets, casted_array_values)
elif isinstance(feature, Sequence):
if feature.length > -1:
if _are_list_values_of_length(array, feature.length):
if array.null_count > 0:
# Ensure each null value in the array translates to [null] * pa_type.list_size in the array's values array
array_type = array.type
storage_type = _storage_type(array_type)
if array_type != storage_type:
# Temporarily convert to the storage type to support extension types in the slice operation
array = array_cast(
array,
storage_type,
allow_primitive_to_str=allow_primitive_to_str,
allow_decimal_to_str=allow_decimal_to_str,
)
array = pc.list_slice(array, 0, feature.length, return_fixed_size_list=True)
array = array_cast(
array,
array_type,
allow_primitive_to_str=allow_primitive_to_str,
allow_decimal_to_str=allow_decimal_to_str,
)
else:
array = pc.list_slice(array, 0, feature.length, return_fixed_size_list=True)
array_values = array.values
casted_array_values = _c(array_values, feature.feature)
return pa.FixedSizeListArray.from_arrays(
casted_array_values, feature.length, mask=array.is_null()
)
else:
array_values = array.values[
array.offset * feature.length : (array.offset + len(array)) * feature.length
]
return pa.FixedSizeListArray.from_arrays(_c(array_values, feature.feature), feature.length)
else:
casted_array_values = _c(array.values, feature.feature)
if pa.types.is_list(array.type) and casted_array_values.type == array.values.type:
# Both array and feature have equal list type and values (within the list) type
return array
else:
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.ListArray.from_arrays(array_offsets, casted_array_values)
elif pa.types.is_fixed_size_list(array.type):
# feature must be either [subfeature] or Sequence(subfeature)
if isinstance(feature, list):
array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
return pa.ListArray.from_arrays(array_offsets, _c(array.values, feature[0]), mask=array.is_null())
elif isinstance(feature, LargeList):
array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
return pa.LargeListArray.from_arrays(
array_offsets, _c(array.values, feature.feature), mask=array.is_null()
)
elif isinstance(feature, Sequence):
if feature.length > -1:
if feature.length == array.type.list_size:
array_values = array.values[
array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
]
casted_array_values = _c(array_values, feature.feature)
return pa.FixedSizeListArray.from_arrays(casted_array_values, feature.length, mask=array.is_null())
else:
array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
return pa.ListArray.from_arrays(array_offsets, _c(array.values, feature.feature), mask=array.is_null())
if pa.types.is_null(array.type):
return array_cast(
array,
get_nested_type(feature),
allow_primitive_to_str=allow_primitive_to_str,
allow_decimal_to_str=allow_decimal_to_str,
)
elif not isinstance(feature, (Sequence, dict, list, tuple)):
return array_cast(
array,
feature(),
allow_primitive_to_str=allow_primitive_to_str,
allow_decimal_to_str=allow_decimal_to_str,
)
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
@_wrap_for_chunked_arrays
def embed_array_storage(array: pa.Array, feature: "FeatureType"):
"""Embed data into an arrays's storage.
For custom features like Audio or Image, it takes into account the "embed_storage" methods
they define to embed external data (e.g. an image file) into an array.
<Added version="2.4.0"/>
Args:
array (`pa.Array`):
The PyArrow array in which to embed data.
feature (`datasets.features.FeatureType`):
Array features.
Raises:
`TypeError`: if the target type is not supported according, e.g.
- if a field is missing
Returns:
array (`pyarrow.Array`): the casted array
"""
from .features import Sequence
_e = embed_array_storage
if isinstance(array, pa.ExtensionArray):
array = array.storage
if hasattr(feature, "embed_storage"):
return feature.embed_storage(array)
elif pa.types.is_struct(array.type):
# feature must be a dict or Sequence(subfeatures_dict)
if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
feature = {
name: Sequence(subfeature, length=feature.length) for name, subfeature in feature.feature.items()
}
if isinstance(feature, dict):
arrays = [_e(array.field(name), subfeature) for name, subfeature in feature.items()]
return pa.StructArray.from_arrays(arrays, names=list(feature), mask=array.is_null())
elif pa.types.is_list(array.type):
# feature must be either [subfeature] or Sequence(subfeature)
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
if isinstance(feature, list):
return pa.ListArray.from_arrays(array_offsets, _e(array.values, feature[0]))
if isinstance(feature, Sequence) and feature.length == -1:
return pa.ListArray.from_arrays(array_offsets, _e(array.values, feature.feature))
elif pa.types.is_large_list(array.type):
# feature must be LargeList(subfeature)
# Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
array_offsets = _combine_list_array_offsets_with_mask(array)
return pa.LargeListArray.from_arrays(array_offsets, _e(array.values, feature.feature))
elif pa.types.is_fixed_size_list(array.type):
# feature must be Sequence(subfeature)
if isinstance(feature, Sequence) and feature.length > -1:
array_values = array.values[
array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
]
embedded_array_values = _e(array_values, feature.feature)
return pa.FixedSizeListArray.from_arrays(embedded_array_values, feature.length, mask=array.is_null())
if not isinstance(feature, (Sequence, dict, list, tuple)):
return array
raise TypeError(f"Couldn't embed array of type\n{_short_str(array.type)}\nwith\n{_short_str(feature)}")
class CastError(ValueError):
"""When it's not possible to cast an Arrow table to a specific schema or set of features"""
def __init__(self, *args, table_column_names: list[str], requested_column_names: list[str]) -> None:
super().__init__(*args)
self.table_column_names = table_column_names
self.requested_column_names = requested_column_names
def __reduce__(self):
# Fix unpickling: TypeError: __init__() missing 2 required keyword-only arguments: 'table_column_names' and 'requested_column_names'
return partial(
CastError, table_column_names=self.table_column_names, requested_column_names=self.requested_column_names
), ()
def details(self):
new_columns = set(self.table_column_names) - set(self.requested_column_names)
missing_columns = set(self.requested_column_names) - set(self.table_column_names)
if new_columns and missing_columns:
return f"there are {len(new_columns)} new columns ({_short_str(new_columns)}) and {len(missing_columns)} missing columns ({_short_str(missing_columns)})."
elif new_columns:
return f"there are {len(new_columns)} new columns ({_short_str(new_columns)})"
else:
return f"there are {len(missing_columns)} missing columns ({_short_str(missing_columns)})"
def cast_table_to_features(table: pa.Table, features: "Features"):
"""Cast a table to the arrow schema that corresponds to the requested features.
Args:
table (`pyarrow.Table`):
PyArrow table to cast.
features ([`Features`]):
Target features.
Returns:
table (`pyarrow.Table`): the casted table
"""
if sorted(table.column_names) != sorted(features):
raise CastError(
f"Couldn't cast\n{_short_str(table.schema)}\nto\n{_short_str(features)}\nbecause column names don't match",
table_column_names=table.column_names,
requested_column_names=list(features),
)
arrays = [cast_array_to_feature(table[name], feature) for name, feature in features.items()]
return pa.Table.from_arrays(arrays, schema=features.arrow_schema)
def cast_table_to_schema(table: pa.Table, schema: pa.Schema):
"""Cast a table to the arrow schema. Different from `cast_table_to_features`, this method can preserve nullability.
Args:
table (`pa.Table`):
PyArrow table to cast.
features ([`Features`]):
Target features.
Returns:
`pa.Table`: the casted table
"""
from .features import Features
features = Features.from_arrow_schema(schema)
table_column_names = set(table.column_names)
if not table_column_names <= set(schema.names):
raise CastError(
f"Couldn't cast\n{_short_str(table.schema)}\nto\n{_short_str(features)}\nbecause column names don't match",
table_column_names=table.column_names,
requested_column_names=list(features),
)
arrays = [
cast_array_to_feature(
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
feature,
)
for name, feature in features.items()
]
return pa.Table.from_arrays(arrays, schema=schema)
def embed_table_storage(table: pa.Table):
"""Embed external data into a table's storage.
<Added version="2.4.0"/>
Args:
table (`pyarrow.Table`):
PyArrow table in which to embed data.
Returns:
table (`pyarrow.Table`): the table with embedded data
"""
from .features.features import Features, require_storage_embed
features = Features.from_arrow_schema(table.schema)
arrays = [
embed_array_storage(table[name], feature) if require_storage_embed(feature) else table[name]
for name, feature in features.items()
]
return pa.Table.from_arrays(arrays, schema=features.arrow_schema)
def table_cast(table: pa.Table, schema: pa.Schema):
"""Improved version of `pa.Table.cast`.
It supports casting to feature types stored in the schema metadata.
Args:
table (`pyarrow.Table`):
PyArrow table to cast.
schema (`pyarrow.Schema`):
Target PyArrow schema.
Returns:
table (`pyarrow.Table`): the casted table
"""
if table.schema != schema:
return cast_table_to_schema(table, schema)
elif table.schema.metadata != schema.metadata:
return table.replace_schema_metadata(schema.metadata)
else:
return table
def table_flatten(table: pa.Table):
"""Improved version of `pa.Table.flatten`.
It behaves as `pa.Table.flatten` in a sense it does 1-step flatten of the columns with a struct type into one column per struct field,
but updates the metadata and skips decodable features unless the `decode` attribute of these features is set to False.
Args:
table (`pa.Table`):
PyArrow table to flatten.
Returns:
`Table`: the flattened table
"""
from .features import Features
features = Features.from_arrow_schema(table.schema)
if any(hasattr(subfeature, "flatten") and subfeature.flatten() == subfeature for subfeature in features.values()):
flat_arrays = []
flat_column_names = []
for field in table.schema:
array = table.column(field.name)
subfeature = features[field.name]
if pa.types.is_struct(field.type) and (
not hasattr(subfeature, "flatten") or subfeature.flatten() != subfeature
):
flat_arrays.extend(array.flatten())
flat_column_names.extend([f"{field.name}.{subfield.name}" for subfield in field.type])
else:
flat_arrays.append(array)
flat_column_names.append(field.name)
flat_table = pa.Table.from_arrays(
flat_arrays,
names=flat_column_names,
)
else:
flat_table = table.flatten()
# Preserve complex types in the metadata
flat_features = features.flatten(max_depth=2)
flat_features = Features({column_name: flat_features[column_name] for column_name in flat_table.column_names})
return flat_table.replace_schema_metadata(flat_features.arrow_schema.metadata)
def table_visitor(table: pa.Table, function: Callable[[pa.Array], None]):
"""Visit all arrays in a table and apply a function to them.
Args:
table (`pyarrow.Table`):
PyArrow table to visit.
function (`Callable[[pa.Array], None]`):
Function to apply to each array.
"""
from .features import Features, Sequence
features = Features.from_arrow_schema(table.schema)
def _visit(array, feature):
if isinstance(array, pa.ChunkedArray):
for chunk in array.chunks:
_visit(chunk, feature)
else:
if isinstance(array, pa.ExtensionArray):
array = array.storage
function(array, feature)
if pa.types.is_struct(array.type) and not hasattr(feature, "cast_storage"):
if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
feature = {
name: Sequence(subfeature, length=feature.length)
for name, subfeature in feature.feature.items()
}
for name, subfeature in feature.items():
_visit(array.field(name), subfeature)
elif pa.types.is_list(array.type):
if isinstance(feature, list):
_visit(array.values, feature[0])
elif isinstance(feature, Sequence):
_visit(array.values, feature.feature)
for name, feature in features.items():
_visit(table[name], feature)
def table_iter(table: Table, batch_size: int, drop_last_batch=False) -> Iterator[pa.Table]:
"""Iterate over sub-tables of size `batch_size`.
Args:
table (`pyarrow.Table`):
PyArrow table to iterate over.
batch_size (`int`):
Size of each sub-table to yield.
drop_last_batch (`bool`, defaults to `False`):
Drop the last batch if it is smaller than `batch_size`.
"""
chunks_buffer = []
chunks_buffer_size = 0
for chunk in table.to_reader(max_chunksize=batch_size):
if len(chunk) == 0:
continue
elif chunks_buffer_size + len(chunk) < batch_size:
chunks_buffer.append(chunk)
chunks_buffer_size += len(chunk)
continue
elif chunks_buffer_size + len(chunk) == batch_size:
chunks_buffer.append(chunk)
yield pa.Table.from_batches(chunks_buffer)
chunks_buffer = []
chunks_buffer_size = 0
else:
cropped_chunk_length = batch_size - chunks_buffer_size
chunks_buffer.append(chunk.slice(0, cropped_chunk_length))
yield pa.Table.from_batches(chunks_buffer)
chunks_buffer = [chunk.slice(cropped_chunk_length, len(chunk) - cropped_chunk_length)]
chunks_buffer_size = len(chunk) - cropped_chunk_length
if not drop_last_batch and chunks_buffer:
yield pa.Table.from_batches(chunks_buffer)
|