Spaces:
Running
Running
File size: 50,495 Bytes
3326079 3a4f3dc 3326079 3a4f3dc 3326079 3a4f3dc 3326079 3a4f3dc 3326079 3a4f3dc 3326079 3a4f3dc 3326079 | 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 | """Pydantic models for structure-aware document chunking.
This module defines the data models used throughout the chunking pipeline.
These models provide type-safe, validated representations of document chunks
and chunking configuration parameters.
Models:
- Chunk: Represents a single chunk of document content with metadata
- ChunkingConfig: Configuration parameters for the chunking process
- TextNormalizer: Utility class for normalizing extracted text
Design Principles:
- All models use Pydantic v2 for validation and serialization
- Comprehensive validation ensures data integrity
- Text normalization handles common PDF extraction artifacts
- JSON/JSONL serialization is optimized for storage and transfer
Lazy Loading:
Pydantic is a lightweight dependency that loads quickly. No lazy loading
is needed for this module.
Example:
-------
>>> from rag_chatbot.chunking.models import Chunk, ChunkingConfig
>>> config = ChunkingConfig(min_tokens=450, max_tokens=700)
>>> chunk = Chunk(
... chunk_id="doc1_chunk_001",
... text="The PMV model predicts thermal sensation...",
... heading_path=["H1: Thermal Comfort", "H2: PMV Model"],
... source="ashrae_55.pdf",
... page=5,
... start_char=1024,
... end_char=2048,
... token_count=156,
... )
>>> print(chunk.chunk_hash) # Auto-generated
'a3f2b1c4d5e6f789'
"""
from __future__ import annotations
import hashlib
import re
from typing import TYPE_CHECKING, Any
from pydantic import (
BaseModel,
ConfigDict,
Field,
field_validator,
model_validator,
)
# =============================================================================
# Type Checking Imports
# =============================================================================
# These imports are only processed by type checkers (mypy, pyright) and IDEs.
# They enable proper type hints without runtime overhead.
# =============================================================================
if TYPE_CHECKING:
from types import ModuleType
from typing import Self
# =============================================================================
# Module Exports
# =============================================================================
__all__: list[str] = [
"Chunk",
"ChunkingConfig",
"TextNormalizer",
"THERMAL_COMFORT_TERMS",
]
# =============================================================================
# Constants
# =============================================================================
# Domain-specific vocabulary for thermal comfort terminology.
# This dictionary maps common variations and OCR errors to their correct forms.
# Used by TextNormalizer to fix jumbled or incorrectly extracted terms.
# The dictionary includes:
# - Standard thermal comfort indices (PMV, PPD, SET, etc.)
# - ASHRAE terminology and standards references
# - Physical units and measurement terms (clo, met, etc.)
# - Common thermal comfort parameters
# - pythermalcomfort library function names
THERMAL_COMFORT_TERMS: dict[str, str] = {
# Thermal comfort indices and models
"p m v": "PMV",
"pmv": "PMV",
"p p d": "PPD",
"ppd": "PPD",
"p m v - p p d": "PMV-PPD",
"pmv-ppd": "PMV-PPD",
"pmv ppd": "PMV-PPD",
"s e t": "SET",
"set*": "SET*",
"s e t *": "SET*",
"u t c i": "UTCI",
"utci": "UTCI",
"p h s": "PHS",
"phs": "PHS",
"a t h b": "ATHB",
"athb": "ATHB",
"adaptive": "adaptive",
"a d a p t i v e": "adaptive",
# Standards and organizations
"a s h r a e": "ASHRAE",
"ashrae": "ASHRAE",
"ashrae 55": "ASHRAE 55",
"ashrae-55": "ASHRAE 55",
"a s h r a e 55": "ASHRAE 55",
"i s o": "ISO",
"iso": "ISO",
"i s o 7730": "ISO 7730",
"iso 7730": "ISO 7730",
"iso-7730": "ISO 7730",
"i s o 7243": "ISO 7243",
"iso 7243": "ISO 7243",
"e n 15251": "EN 15251",
"en 15251": "EN 15251",
"en-15251": "EN 15251",
"e n 16798": "EN 16798",
"en 16798": "EN 16798",
"en-16798": "EN 16798",
# Physical units and measurements
"c l o": "clo",
"m e t": "met",
"w / m 2": "W/m2",
"w/m2": "W/m2",
"w / m^2": "W/m2",
"w/m^2": "W/m2",
"m / s": "m/s",
"k p a": "kPa",
"kpa": "kPa",
"p a": "Pa",
"deg c": "degC",
"deg f": "degF",
"deg k": "degK",
# Thermal comfort parameters
"t a": "ta", # air temperature
"t r": "tr", # radiant temperature
"t o": "to", # operative temperature
"t o p": "top", # operative temperature
"t m r t": "tmrt", # mean radiant temperature
"tmrt": "tmrt",
"t mrt": "tmrt",
"m r t": "MRT", # mean radiant temperature
"mrt": "MRT",
"r h": "RH", # relative humidity
"v e l": "vel", # velocity
"v a": "va", # air velocity
"v r": "vr", # relative air velocity
# pythermalcomfort specific terms
"pythermalcomfort": "pythermalcomfort",
"p y t h e r m a l c o m f o r t": "pythermalcomfort",
"thermal comfort": "thermal comfort",
"ther mal com fort": "thermal comfort",
"ther mal": "thermal",
"com fort": "comfort",
# Common function names from the library (preserving underscores)
"pmv_ppd": "pmv_ppd",
"adaptive_ashrae": "adaptive_ashrae",
"adaptive_en": "adaptive_en",
"clo_dynamic": "clo_dynamic",
"cooling_effect": "cooling_effect",
"set_tmp": "set_tmp",
"solar_gain": "solar_gain",
"use_fans_heatwaves": "use_fans_heatwaves",
"wbgt": "WBGT",
"w b g t": "WBGT",
"heat_index": "heat_index",
"humidex": "humidex",
"net": "NET",
"n e t": "NET",
"at": "AT", # apparent temperature
"a t": "AT",
"wind_chill": "wind_chill",
"phs_model": "phs_model",
"two_nodes": "two_nodes",
"solar_altitude": "solar_altitude",
"mean_radiant_temperature": "mean_radiant_temperature",
# =========================================================================
# pythermalcomfort function name corrections
# Maps concatenated versions (without underscores) to correct snake_case
# This handles cases where PDF extraction strips underscores from names.
# =========================================================================
# Models - PMV/PPD variants
"pmvppdashrae": "pmv_ppd_ashrae",
"pmvppdiso": "pmv_ppd_iso",
"pmvathb": "pmv_athb",
"pmva": "pmv_a",
"pmve": "pmv_e",
# Models - Adaptive comfort
"adaptiveashrae": "adaptive_ashrae",
"adaptiveen": "adaptive_en",
# Models - Two-node models
"twonodesgagge": "two_nodes_gagge",
"twonodesgaggesleep": "two_nodes_gagge_sleep",
"twonodesgaggeji": "two_nodes_gagge_ji",
# Models - Heat indices
"heatindexlu": "heat_index_lu",
"heatindexrothfusz": "heat_index_rothfusz",
"discomfortindex": "discomfort_index",
# Models - Other thermal indices
"petsteady": "pet_steady",
"settmp": "set_tmp",
"coolingeffect": "cooling_effect",
"solargain": "solar_gain",
"usefansheatwaves": "use_fans_heatwaves",
"verticaltmpgradppd": "vertical_tmp_grad_ppd",
"ankledraft": "ankle_draft",
"clotout": "clo_tout",
# Models - Work capacity
"workcapacitydunne": "work_capacity_dunne",
"workcapacityhothaps": "work_capacity_hothaps",
"workcapacityiso": "work_capacity_iso",
"workcapacityniosh": "work_capacity_niosh",
# Models - Wind chill
"windchilltemperature": "wind_chill_temperature",
# Utilities - Temperature and psychrometrics
"runningmeanoutdoortemperature": "running_mean_outdoor_temperature",
"meanradianttmp": "mean_radiant_tmp",
"operativetmp": "operative_tmp",
"dewpointtmp": "dew_point_tmp",
"wetbulbtmp": "wet_bulb_tmp",
"enthalpyair": "enthalpy_air",
"bodysurfacearea": "body_surface_area",
"psytarh": "psy_ta_rh",
"vrelative": "v_relative",
"unitsconverter": "units_converter",
# Utilities - Clothing functions
"clodynamicashrae": "clo_dynamic_ashrae",
"clodynamiciso": "clo_dynamic_iso",
"cloinsulationairlayer": "clo_insulation_air_layer",
"cloareafactor": "clo_area_factor",
"clocorrectionfactorenvironment": "clo_correction_factor_environment",
"clointrinsicinsulatioensemble": "clo_intrinsic_insulation_ensemble",
"clototalinsulation": "clo_total_insulation",
"clotypicalensembles": "clo_typical_ensembles",
"cloindividualgarments": "clo_individual_garments",
"mettypicaltasks": "met_typical_tasks",
}
# Regex pattern for detecting ALL CAPS words (3+ consecutive capital letters)
# Used for heading normalization
_ALL_CAPS_PATTERN: re.Pattern[str] = re.compile(r"\b([A-Z]{3,})\b")
# Regex pattern for detecting mid-word spaces (common OCR artifact)
# Matches single letters separated by spaces: "t h e r m a l" -> "thermal"
_JUMBLED_WORD_PATTERN: re.Pattern[str] = re.compile(r"\b([a-zA-Z])\s+(?=[a-zA-Z]\b)")
# Regex pattern for fixing missing spaces after punctuation
# Matches period/comma/semicolon followed immediately by a letter
_MISSING_SPACE_PATTERN: re.Pattern[str] = re.compile(r"([.!?,;:])([A-Za-z])")
# Regex pattern for multiple consecutive whitespace characters
_MULTI_WHITESPACE_PATTERN: re.Pattern[str] = re.compile(r"[ \t]+")
# Regex pattern for multiple consecutive newlines (more than 2)
_MULTI_NEWLINE_PATTERN: re.Pattern[str] = re.compile(r"\n{3,}")
# Threshold for considering text as "predominantly uppercase" (80%)
# Used in capitalization normalization to detect ALL CAPS text
_UPPERCASE_THRESHOLD: float = 0.8
# Minimum length for an acronym (e.g., "PMV" has length 3, "AT" has length 2)
_MIN_ACRONYM_LENGTH: int = 2
# Minimum length of a word to consider for segmentation
# Words shorter than this are unlikely to be concatenated
# Set to 12 to catch common OCR errors like "conditionsthat" (14 chars)
_MIN_SEGMENT_WORD_LENGTH: int = 12
# Maximum length of a single word in English (reasonable limit)
# Words longer than this are almost certainly concatenated
_MAX_SINGLE_WORD_LENGTH: int = 25
# Minimum length for a valid word segment
# Segments shorter than this are likely errors (e.g., single letters)
_MIN_VALID_SEGMENT_LENGTH: int = 2
# Regex pattern to match HTML comments (e.g., <!-- Page 4 -->)
# These are added by ExtractedDocument.to_markdown() to mark page boundaries
# but should be stripped before creating chunks for embedding
_HTML_COMMENT_PATTERN: re.Pattern[str] = re.compile(r"<!--.*?-->", re.DOTALL)
# Technical terms that should NOT be segmented
# These are valid compound words or domain-specific terms
#
# IMPORTANT: This list includes pythermalcomfort function names in their
# concatenated form (without underscores) because PDF extraction sometimes
# strips underscores. When a word like "pmvppdashrae" is encountered, it
# should NOT be segmented into "pmv ppd ashrae" - instead, it should be
# preserved so that downstream processing or the LLM can recognise it as
# a function name variant.
#
# The function names are extracted from:
# pythermalcomfort-readthedocs-io-en-latest.pdf (official documentation)
_PROTECTED_TERMS: frozenset[str] = frozenset(
{
# General technical terms
"pythermalcomfort",
"thermalcomfort",
"metabolicrate",
"ashrae",
"coefficient",
"coefficients",
"environmental",
"physiological",
"temperature",
"temperatures",
# =====================================================================
# pythermalcomfort.models function names (concatenated, lowercase)
# These protect against incorrect segmentation of function names
# when underscores are stripped during PDF extraction.
# =====================================================================
"adaptiveashrae", # adaptive_ashrae
"adaptiveen", # adaptive_en
"ankledraft", # ankle_draft
"clotout", # clo_tout
"coolingeffect", # cooling_effect
"discomfortindex", # discomfort_index
"twonodesgagge", # two_nodes_gagge
"twonodesgaggesleep", # two_nodes_gagge_sleep
"twonodesgaggeji", # two_nodes_gagge_ji
"heatindexlu", # heat_index_lu
"heatindexrothfusz", # heat_index_rothfusz
"petsteady", # pet_steady
"pmvppdiso", # pmv_ppd_iso
"pmvppdashrae", # pmv_ppd_ashrae
"pmvathb", # pmv_athb
"solargain", # solar_gain
"settmp", # set_tmp
"usefansheatwaves", # use_fans_heatwaves
"verticaltmpgradppd", # vertical_tmp_grad_ppd
"windchilltemperature", # wind_chill_temperature
"workcapacitydunne", # work_capacity_dunne
"workcapacityhothaps", # work_capacity_hothaps
"workcapacityiso", # work_capacity_iso
"workcapacityniosh", # work_capacity_niosh
# =====================================================================
# pythermalcomfort.utilities function names (concatenated, lowercase)
# =====================================================================
"runningmeanoutdoortemperature", # running_mean_outdoor_temperature
"vrelative", # v_relative
"clodynamicashrae", # clo_dynamic_ashrae
"clodynamiciso", # clo_dynamic_iso
"bodysurfacearea", # body_surface_area
"dewpointtmp", # dew_point_tmp
"enthalpyair", # enthalpy_air
"meanradianttmp", # mean_radiant_tmp
"operativetmp", # operative_tmp
"psytarh", # psy_ta_rh
"psat", # p_sat
"fsvv", # f_svv
"unitsconverter", # units_converter
"wetbulbtmp", # wet_bulb_tmp
"cloinsulationairlayer", # clo_insulation_air_layer
"cloareafactor", # clo_area_factor
"clocorrectionfactorenvironment", # clo_correction_factor_environment
"clointrinsicinsulatioensemble", # clo_intrinsic_insulation_ensemble
"clototalinsulation", # clo_total_insulation
"clotypicalensembles", # clo_typical_ensembles
"cloindividualgarments", # clo_individual_garments
"mettypicaltasks", # met_typical_tasks
}
)
# =============================================================================
# Lazy Loading for wordsegment
# =============================================================================
# The wordsegment library is used for detecting and segmenting concatenated
# words from OCR/PDF extraction errors. It is lazily loaded to avoid import
# overhead when not needed.
# =============================================================================
# Lazy-loaded wordsegment module
_wordsegment_module: ModuleType | None = None
_wordsegment_loaded: bool = False
def _get_wordsegment() -> ModuleType:
"""Lazily load and initialize the wordsegment module.
This function loads the wordsegment library on first use and initializes
its word frequency data. Subsequent calls return the cached module.
Returns:
-------
The initialized wordsegment module.
Note:
----
The wordsegment.load() call must happen once before using segment().
This function handles that initialization automatically.
"""
global _wordsegment_module, _wordsegment_loaded # noqa: PLW0603
if not _wordsegment_loaded:
import wordsegment
wordsegment.load() # Load word frequency data
_wordsegment_module = wordsegment
_wordsegment_loaded = True
# At this point _wordsegment_module is guaranteed to be set
assert _wordsegment_module is not None
return _wordsegment_module
# =============================================================================
# Text Normalization
# =============================================================================
class TextNormalizer:
"""Utility class for normalizing extracted text from PDF documents.
PDF extraction often introduces artifacts such as:
- Extra whitespace from column layouts
- Mid-word spaces from OCR errors (e.g., "ther mal" instead of "thermal")
- ALL CAPS headings that should be title case
- Missing spaces after punctuation
This class provides methods to fix these common issues while preserving
the semantic content of the text.
The normalizer uses a domain-specific dictionary of thermal comfort terms
to correctly handle specialized vocabulary that might be incorrectly
extracted by OCR or PDF parsing.
Attributes:
----------
domain_terms : dict[str, str]
Dictionary mapping incorrect/jumbled terms to their correct forms.
Defaults to THERMAL_COMFORT_TERMS if not provided.
Example:
-------
>>> normalizer = TextNormalizer()
>>> text = "The ther mal com fort index"
>>> normalizer.normalize(text)
'The thermal comfort index'
Note:
----
Normalization is applied in a specific order to avoid conflicts:
1. Whitespace normalization (basic cleanup)
2. Jumbled word fixing (domain-specific)
3. Capitalization fixing (headings only)
4. Sentence spacing (punctuation)
"""
def __init__(
self,
domain_terms: dict[str, str] | None = None,
) -> None:
"""Initialize the text normalizer with optional domain terms.
Args:
----
domain_terms: Optional dictionary mapping incorrect terms to their
correct forms. If not provided, uses THERMAL_COMFORT_TERMS.
Example:
-------
>>> normalizer = TextNormalizer()
>>> custom_normalizer = TextNormalizer({"cust om": "custom"})
"""
# Use provided domain terms or fall back to thermal comfort vocabulary
self.domain_terms: dict[str, str] = (
domain_terms if domain_terms is not None else THERMAL_COMFORT_TERMS
)
# Pre-compile patterns for domain terms for efficient matching
# Sort by length (longest first) to match longer phrases before shorter ones
self._term_patterns: list[tuple[re.Pattern[str], str]] = []
for incorrect, correct in sorted(
self.domain_terms.items(),
key=lambda x: len(x[0]),
reverse=True,
):
# Use word boundaries to avoid partial matches
# Case-insensitive matching for flexibility
pattern = re.compile(
r"\b" + re.escape(incorrect) + r"\b",
re.IGNORECASE,
)
self._term_patterns.append((pattern, correct))
def normalize_whitespace(self, text: str) -> str:
r"""Normalize whitespace in the text.
This method:
- Replaces multiple spaces/tabs with a single space
- Normalizes line endings to Unix-style (LF)
- Reduces multiple blank lines to at most two
- Strips leading/trailing whitespace
Args:
----
text: The text to normalize.
Returns:
-------
Text with normalized whitespace.
Example:
-------
>>> normalizer = TextNormalizer()
>>> text = "Hello world\r\n\r\n\r\nTest"
>>> normalizer.normalize_whitespace(text)
'Hello world\n\nTest'
"""
# Return early for empty strings
if not text:
return text
# Normalize line endings: CRLF -> LF, CR -> LF
result = text.replace("\r\n", "\n").replace("\r", "\n")
# Replace multiple spaces/tabs with single space (preserve newlines)
result = _MULTI_WHITESPACE_PATTERN.sub(" ", result)
# Reduce multiple newlines (3+) to double newlines (paragraph break)
result = _MULTI_NEWLINE_PATTERN.sub("\n\n", result)
# Strip leading/trailing whitespace
return result.strip()
def normalize_jumbled_words(self, text: str) -> str:
"""Fix mid-word spaces caused by OCR or PDF extraction.
This method uses the domain dictionary to fix common jumbled words
in thermal comfort terminology. It handles cases like:
- "ther mal" -> "thermal"
- "p m v" -> "PMV"
- "pythermalcomfort" (various broken forms)
Args:
----
text: The text containing potentially jumbled words.
Returns:
-------
Text with jumbled words fixed according to the domain dictionary.
Example:
-------
>>> normalizer = TextNormalizer()
>>> text = "Calculate p m v using pythermalcomfort"
>>> normalizer.normalize_jumbled_words(text)
'Calculate PMV using pythermalcomfort'
Note:
----
This method only fixes terms that are in the domain dictionary.
Unknown jumbled words are left unchanged to avoid false corrections.
"""
if not text:
return text
result = text
# Apply domain-specific corrections using pre-compiled patterns
for pattern, replacement in self._term_patterns:
result = pattern.sub(replacement, result)
return result
def normalize_capitalization(self, text: str, is_heading: bool = False) -> str:
"""Normalize capitalization, especially for ALL CAPS text.
For headings, this method converts ALL CAPS text to Title Case
while preserving known acronyms (PMV, ASHRAE, etc.) from the
domain dictionary.
For regular text, only obvious ALL CAPS sentences are converted.
Args:
----
text: The text to normalize.
is_heading: If True, apply more aggressive title case conversion.
Returns:
-------
Text with normalized capitalization.
Example:
-------
>>> normalizer = TextNormalizer()
>>> normalizer.normalize_capitalization("THERMAL COMFORT", is_heading=True)
'Thermal Comfort'
>>> normalizer.normalize_capitalization("PMV INDEX", is_heading=True)
'PMV Index'
Note:
----
Acronyms from the domain dictionary are preserved in their
correct form (e.g., "PMV" stays as "PMV", not "Pmv").
"""
if not text:
return text
# For non-headings, only process if the entire text is uppercase
if not is_heading:
# Check if text is predominantly uppercase (>80% capital letters)
alpha_chars = [c for c in text if c.isalpha()]
if not alpha_chars:
return text
upper_ratio = sum(1 for c in alpha_chars if c.isupper()) / len(alpha_chars)
if upper_ratio < _UPPERCASE_THRESHOLD:
return text
# Convert to title case
result = text.title()
# Restore known acronyms from domain dictionary
# These should remain in their original uppercase form
acronyms = {
v
for v in self.domain_terms.values()
if v.isupper() and len(v) >= _MIN_ACRONYM_LENGTH
}
for acronym in acronyms:
# Replace title-cased version with correct acronym
title_version = acronym.title()
if title_version in result:
result = result.replace(title_version, acronym)
return result
def normalize_sentences(self, text: str) -> str:
"""Fix missing spaces after punctuation marks.
PDF extraction sometimes loses spaces after periods, commas,
and other punctuation marks. This method adds them back.
Args:
----
text: The text with potential missing spaces.
Returns:
-------
Text with proper spacing after punctuation.
Example:
-------
>>> normalizer = TextNormalizer()
>>> normalizer.normalize_sentences("First sentence.Second sentence")
'First sentence. Second sentence'
Note:
----
This only adds spaces after punctuation when followed directly
by a letter. It won't affect abbreviations like "Dr." when
properly spaced.
"""
if not text:
return text
# Add space after punctuation marks when followed by a letter
result = _MISSING_SPACE_PATTERN.sub(r"\1 \2", text)
return result
def segment_concatenated_words(self, text: str) -> str:
"""Segment concatenated words that are missing spaces.
Uses the wordsegment library to detect and split words that
appear to be concatenated due to OCR or PDF extraction errors.
Examples:
--------
- "conditionsthat" -> "conditions that"
- "theenvironment" -> "the environment"
Args:
----
text: The text containing potential concatenated words.
Returns:
-------
Text with concatenated words segmented.
Note:
----
- Only processes words longer than _MIN_SEGMENT_WORD_LENGTH
- Preserves protected terms from _PROTECTED_TERMS
- Preserves words in the domain dictionary
"""
if not text:
return text
ws = _get_wordsegment()
# Split into words while preserving punctuation and whitespace
words = text.split()
result_words: list[str] = []
# Define punctuation characters to strip (avoid escaping issues)
punct_chars = ".,;:!?()[]{}\"'"
for word in words:
# Strip punctuation for checking, but preserve it
stripped = word.strip(punct_chars)
punct_start = word[: len(word) - len(word.lstrip(punct_chars))]
punct_end = word[len(word.rstrip(punct_chars)) :]
# Skip short words
if len(stripped) < _MIN_SEGMENT_WORD_LENGTH:
result_words.append(word)
continue
# Skip words containing underscores - these are Python identifiers
# (e.g., pmv_ppd_ashrae, clo_dynamic_iso) that should be preserved
# exactly as-is. Underscores in function names are intentional and
# segmenting them would corrupt the identifier.
if "_" in stripped:
result_words.append(word)
continue
# Skip protected terms (case-insensitive)
if stripped.lower() in _PROTECTED_TERMS:
result_words.append(word)
continue
# Skip words in domain dictionary (case-insensitive)
if stripped.lower() in {k.lower() for k in self.domain_terms}:
result_words.append(word)
continue
if stripped.lower() in {v.lower() for v in self.domain_terms.values()}:
result_words.append(word)
continue
# Skip words that look like they're already properly spaced
# (contain uppercase in the middle, suggesting camelCase/acronyms)
if any(c.isupper() for c in stripped[1:-1] if c.isalpha()):
result_words.append(word)
continue
# Segment the word using wordsegment library
segments: list[str] = ws.segment(stripped.lower())
# Only accept segmentation if it produces multiple reasonable words
if len(segments) > 1 and all(
len(s) >= _MIN_VALID_SEGMENT_LENGTH for s in segments
):
# Preserve original capitalization of first letter if uppercase
if stripped[0].isupper():
segments[0] = segments[0].capitalize()
segmented = " ".join(segments)
result_words.append(punct_start + segmented + punct_end)
else:
result_words.append(word)
return " ".join(result_words)
def strip_html_comments(self, text: str) -> str:
"""Remove HTML comments from text.
PDF extraction adds HTML comments like `<!-- Page 4 -->` to mark
page boundaries. These should be stripped before creating chunks
for embedding, as they add noise without semantic value.
Args:
----
text: The text potentially containing HTML comments.
Returns:
-------
Text with all HTML comments removed.
Example:
-------
>>> normalizer = TextNormalizer()
>>> text = "Hello <!-- Page 1 --> world <!-- Page 2 -->"
>>> normalizer.strip_html_comments(text)
'Hello world '
Note:
----
This method removes all HTML comments, not just page markers.
The surrounding whitespace may need cleanup after removal,
which is handled by subsequent whitespace normalization.
"""
if not text:
return text
return _HTML_COMMENT_PATTERN.sub("", text)
def normalize(self, text: str, is_heading: bool = False) -> str:
"""Apply all normalizations to the text.
This is the main method that combines all normalization steps
in the correct order:
0. HTML comment stripping (remove page markers)
1. Whitespace normalization (basic cleanup)
2. Jumbled word fixing (domain-specific corrections)
2.5. Concatenated word segmentation (OCR missing spaces)
3. Sentence spacing (punctuation fixes)
4. Capitalization (if needed for headings)
Args:
----
text: The text to normalize.
is_heading: If True, apply heading-specific normalization
(e.g., title case for ALL CAPS).
Returns:
-------
Fully normalized text.
Example:
-------
>>> normalizer = TextNormalizer()
>>> text = "The ther mal com fort index.Is used for"
>>> normalizer.normalize(text)
'The thermal comfort index. Is used for'
"""
if not text:
return text
# Step 0: Strip HTML comments (page markers, etc.)
result = self.strip_html_comments(text)
# Step 1: Normalize whitespace first (basic cleanup)
result = self.normalize_whitespace(result)
# Step 2: Fix jumbled words using domain dictionary
result = self.normalize_jumbled_words(result)
# Step 2.5: Segment concatenated words (missing spaces from OCR)
result = self.segment_concatenated_words(result)
# Step 3: Fix missing spaces after punctuation
result = self.normalize_sentences(result)
# Step 4: Normalize capitalization (especially for headings)
if is_heading:
result = self.normalize_capitalization(result, is_heading=True)
return result
# =============================================================================
# Data Models
# =============================================================================
class Chunk(BaseModel):
"""Represent a single chunk of document content with metadata.
A chunk is a semantically meaningful portion of a document that is
suitable for embedding and retrieval. Each chunk contains:
- The text content itself
- Heading hierarchy for context (e.g., ["H1: Chapter", "H2: Section"])
- Source file and page information
- Character offsets for traceability
- Token count for size management
- Content hash for deduplication
The chunk model is designed to be serialized to JSONL format for
storage and loading during the retrieval pipeline.
Attributes:
----------
chunk_id : str
Unique identifier for the chunk within the corpus.
Typically formatted as "{source}_{index}" for traceability.
text : str
The actual text content of the chunk. This is the content
that will be embedded and retrieved.
heading_path : list[str]
Hierarchical list of headings providing context.
Format: ["H1: Title", "H2: Section", "H3: Subsection"]
Empty list if no heading hierarchy is available.
source : str
The original filename or source document identifier.
Used for citation and source attribution.
page : int
The 1-indexed page number where this chunk originates.
Must be >= 1 as PDF pages are conventionally numbered from 1.
start_char : int
Starting character offset in the source document.
Must be >= 0.
end_char : int
Ending character offset in the source document (exclusive).
Must be > start_char.
token_count : int
Approximate number of tokens in the chunk text.
Used for managing chunk sizes during retrieval.
chunk_hash : str
SHA-256 hash of the text content (first 16 characters).
Auto-generated if not provided. Used for deduplication.
Example:
-------
>>> chunk = Chunk(
... chunk_id="ashrae55_001",
... text="The PMV model predicts thermal sensation...",
... heading_path=["H1: Thermal Comfort", "H2: PMV Model"],
... source="ashrae_55.pdf",
... page=5,
... start_char=1024,
... end_char=2048,
... token_count=156,
... )
>>> chunk.chunk_hash # Auto-generated
'a3f2b1c4d5e6f789'
>>> chunk.text_preview(50)
'The PMV model predicts thermal sensation...'
Note:
----
The chunk_hash is automatically generated from the text content
if not explicitly provided. This enables efficient deduplication
and change detection.
"""
# -------------------------------------------------------------------------
# Model Configuration
# -------------------------------------------------------------------------
# Configure Pydantic model behavior for serialization and validation
# -------------------------------------------------------------------------
model_config = ConfigDict(
# Allow population by field name or alias
populate_by_name=True,
# Validate default values during model creation
validate_default=True,
# Use enum values in serialization rather than enum objects
use_enum_values=True,
# Extra fields are forbidden to catch typos and ensure data integrity
extra="forbid",
# Enable JSON schema generation with examples
json_schema_extra={
"examples": [
{
"chunk_id": "ashrae55_001",
"text": "The PMV model predicts thermal sensation...",
"heading_path": ["H1: Thermal Comfort", "H2: PMV Model"],
"source": "ashrae_55.pdf",
"page": 5,
"start_char": 1024,
"end_char": 2048,
"token_count": 156,
"chunk_hash": "a3f2b1c4d5e6f789",
},
{
"chunk_id": "iso7730_042",
"text": "The PPD index represents the percentage...",
"heading_path": ["H1: ISO 7730", "H2: PPD Calculation"],
"source": "iso_7730.pdf",
"page": 12,
"start_char": 5120,
"end_char": 6144,
"token_count": 189,
"chunk_hash": "b4e3c2d1f5a6e8c7",
},
]
},
)
# -------------------------------------------------------------------------
# Fields
# -------------------------------------------------------------------------
chunk_id: str = Field(
..., # Required field (no default)
min_length=1, # Must not be empty
description="Unique identifier for the chunk within the corpus",
examples=["ashrae55_001", "iso7730_042", "guide_chapter2_015"],
)
text: str = Field(
..., # Required field
min_length=1, # Must not be empty
description="The text content of the chunk",
examples=["The PMV model predicts thermal sensation based on..."],
)
heading_path: list[str] = Field(
default_factory=list,
description="Hierarchical list of headings providing context",
examples=[["H1: Thermal Comfort", "H2: PMV Model", "H3: Calculation"]],
)
source: str = Field(
..., # Required field
min_length=1, # Must not be empty
description="Original filename or source document identifier",
examples=["ashrae_55.pdf", "iso_7730.pdf", "pythermalcomfort_guide.pdf"],
)
page: int = Field(
..., # Required field
ge=1, # Must be >= 1 (1-indexed page numbers)
description="1-indexed page number where this chunk originates",
examples=[1, 5, 42],
)
start_char: int = Field(
..., # Required field
ge=0, # Must be >= 0
description="Starting character offset in the source document",
examples=[0, 1024, 5120],
)
end_char: int = Field(
..., # Required field
gt=0, # Must be > 0 (will be validated further against start_char)
description="Ending character offset in the source document (exclusive)",
examples=[512, 2048, 6144],
)
token_count: int = Field(
..., # Required field
ge=0, # Must be >= 0 (can be 0 for empty-ish chunks)
description="Approximate number of tokens in the chunk text",
examples=[100, 256, 512],
)
chunk_hash: str = Field(
default="", # Will be auto-generated in model_post_init
max_length=16, # SHA-256 truncated to 16 characters
description="SHA-256 hash of text content (first 16 chars) for deduplication",
examples=["a3f2b1c4d5e6f789", "b4e3c2d1f5a6e8c7"],
)
# -------------------------------------------------------------------------
# Validators
# -------------------------------------------------------------------------
@field_validator("text", mode="before")
@classmethod
def _normalize_text(cls, value: object) -> str:
"""Normalize text content and strip leading/trailing whitespace.
Args:
----
value: The input value to normalize.
Returns:
-------
Normalized string content.
Raises:
------
ValueError: If value is None or empty after stripping.
"""
if value is None:
msg = "text cannot be None"
raise ValueError(msg)
text = str(value).strip()
if not text:
msg = "text cannot be empty"
raise ValueError(msg)
return text
@field_validator("heading_path", mode="before")
@classmethod
def _ensure_heading_list(cls, value: object) -> list[str]:
"""Ensure heading_path is always a list of strings.
Args:
----
value: The input value to normalize.
Returns:
-------
List of heading strings.
"""
if value is None:
return []
if isinstance(value, str):
# Single heading provided as string
return [value] if value.strip() else []
if isinstance(value, list):
# Filter out empty strings and convert all to strings
return [str(h).strip() for h in value if str(h).strip()]
# Handle other iterables
try:
iterator = iter(value) # type: ignore[call-overload]
return [str(h).strip() for h in iterator if str(h).strip()]
except TypeError:
# Not iterable, wrap in list if non-empty
h_str = str(value).strip()
return [h_str] if h_str else []
@model_validator(mode="after")
def _validate_char_offsets(self) -> Self:
"""Validate that end_char is greater than start_char.
Returns
-------
The validated model instance.
Raises
------
ValueError: If end_char is not greater than start_char.
"""
if self.end_char <= self.start_char:
msg = (
f"end_char ({self.end_char}) must be greater than "
f"start_char ({self.start_char})"
)
raise ValueError(msg)
return self
# -------------------------------------------------------------------------
# Post-Initialization
# -------------------------------------------------------------------------
def model_post_init(self, __context: object) -> None:
"""Generate chunk_hash if not provided.
This method is called after the model is fully initialized.
It generates a SHA-256 hash of the text content (truncated to
16 characters) if the chunk_hash field was not explicitly set.
Args:
----
__context: Pydantic context object (unused but required by signature).
Note:
----
The hash is deterministic - the same text will always produce
the same hash, enabling deduplication across runs.
"""
# Only generate hash if not already set
if not self.chunk_hash:
# Generate SHA-256 hash of text content
text_bytes = self.text.encode("utf-8")
full_hash = hashlib.sha256(text_bytes).hexdigest()
# Use object.__setattr__ to bypass frozen model if needed
object.__setattr__(self, "chunk_hash", full_hash[:16])
# -------------------------------------------------------------------------
# Methods
# -------------------------------------------------------------------------
def to_jsonl_dict(self) -> dict[str, Any]:
"""Export the chunk to a dictionary suitable for JSONL serialization.
This method produces a flat dictionary representation that can be
written to a JSONL file. All values are JSON-serializable.
Returns:
-------
Dictionary with all chunk fields ready for JSON serialization.
Example:
-------
>>> chunk = Chunk(
... chunk_id="test_001",
... text="Example text",
... source="test.pdf",
... page=1,
... start_char=0,
... end_char=12,
... token_count=2,
... )
>>> data = chunk.to_jsonl_dict()
>>> data["chunk_id"]
'test_001'
Note:
----
This method is preferred over model_dump() for JSONL output
as it ensures consistent field ordering and formatting.
"""
return {
"chunk_id": self.chunk_id,
"text": self.text,
"heading_path": self.heading_path,
"source": self.source,
"page": self.page,
"start_char": self.start_char,
"end_char": self.end_char,
"token_count": self.token_count,
"chunk_hash": self.chunk_hash,
}
def text_preview(self, max_length: int = 100) -> str:
"""Get a truncated preview of the chunk text.
This method returns the first `max_length` characters of the
text, with an ellipsis appended if the text was truncated.
Args:
----
max_length: Maximum number of characters to include.
Defaults to 100.
Returns:
-------
Truncated text with ellipsis if needed.
Example:
-------
>>> chunk = Chunk(
... chunk_id="test_001",
... text="This is a very long text that needs truncation",
... source="test.pdf",
... page=1,
... start_char=0,
... end_char=47,
... token_count=10,
... )
>>> chunk.text_preview(20)
'This is a very lo...'
"""
if len(self.text) <= max_length:
return self.text
# Truncate and add ellipsis
return self.text[: max_length - 3] + "..."
class ChunkingConfig(BaseModel):
"""Configuration parameters for the document chunking process.
This model defines the parameters that control how documents are
split into chunks. The parameters balance several concerns:
- Chunk size (min/max tokens)
- Context overlap between chunks
- Preservation of natural text boundaries
The default values are tuned for thermal comfort documentation
and the BGE embedding model used in this pipeline.
Attributes:
----------
min_tokens : int
Minimum number of tokens per chunk. Chunks smaller than
this will be merged with adjacent content. Default: 450.
max_tokens : int
Maximum number of tokens per chunk. Content exceeding
this limit will be split. Default: 700.
overlap_percent : float
Percentage of max_tokens to overlap between consecutive
chunks. Helps maintain context across chunk boundaries.
Must be between 0.0 and 1.0. Default: 0.12 (12%).
preserve_sentences : bool
If True, avoid splitting in the middle of sentences.
Default: True.
preserve_paragraphs : bool
If True, prefer paragraph boundaries as split points.
Default: True.
Example:
-------
>>> config = ChunkingConfig(min_tokens=400, max_tokens=600)
>>> config.calculate_overlap_tokens()
72
>>> config = ChunkingConfig(overlap_percent=0.15)
>>> config.calculate_overlap_tokens()
105
Note:
----
The overlap is calculated as: overlap_tokens = max_tokens * overlap_percent
For the default values: 700 * 0.12 = 84 tokens of overlap.
"""
# -------------------------------------------------------------------------
# Model Configuration
# -------------------------------------------------------------------------
model_config = ConfigDict(
# Allow population by field name or alias
populate_by_name=True,
# Validate default values during model creation
validate_default=True,
# Extra fields are forbidden
extra="forbid",
# JSON schema examples
json_schema_extra={
"examples": [
{
"min_tokens": 450,
"max_tokens": 700,
"overlap_percent": 0.12,
"preserve_sentences": True,
"preserve_paragraphs": True,
},
{
"min_tokens": 300,
"max_tokens": 512,
"overlap_percent": 0.10,
"preserve_sentences": True,
"preserve_paragraphs": False,
},
]
},
)
# -------------------------------------------------------------------------
# Fields
# -------------------------------------------------------------------------
min_tokens: int = Field(
default=450,
ge=1, # Must be at least 1 token
description="Minimum number of tokens per chunk",
examples=[300, 450, 500],
)
max_tokens: int = Field(
default=700,
ge=1, # Must be at least 1 token
description="Maximum number of tokens per chunk",
examples=[512, 700, 1024],
)
overlap_percent: float = Field(
default=0.12,
ge=0.0, # At least 0% overlap
le=1.0, # At most 100% overlap
description="Percentage of max_tokens to overlap between chunks (0.0-1.0)",
examples=[0.10, 0.12, 0.15, 0.20],
)
preserve_sentences: bool = Field(
default=True,
description="Avoid splitting in the middle of sentences",
)
preserve_paragraphs: bool = Field(
default=True,
description="Prefer paragraph boundaries as split points",
)
# -------------------------------------------------------------------------
# Validators
# -------------------------------------------------------------------------
@model_validator(mode="after")
def _validate_token_range(self) -> Self:
"""Validate that min_tokens is less than max_tokens.
Returns
-------
The validated model instance.
Raises
------
ValueError: If min_tokens is greater than or equal to max_tokens.
"""
if self.min_tokens >= self.max_tokens:
msg = (
f"min_tokens ({self.min_tokens}) must be less than "
f"max_tokens ({self.max_tokens})"
)
raise ValueError(msg)
return self
# -------------------------------------------------------------------------
# Methods
# -------------------------------------------------------------------------
def calculate_overlap_tokens(self) -> int:
"""Calculate the number of overlap tokens based on configuration.
This method computes the actual number of tokens to overlap
between consecutive chunks based on max_tokens and overlap_percent.
Returns:
-------
Number of tokens to overlap between consecutive chunks.
Example:
-------
>>> config = ChunkingConfig(max_tokens=700, overlap_percent=0.12)
>>> config.calculate_overlap_tokens()
84
>>> config = ChunkingConfig(max_tokens=512, overlap_percent=0.10)
>>> config.calculate_overlap_tokens()
51
Note:
----
The result is rounded to the nearest integer.
"""
return round(self.max_tokens * self.overlap_percent)
|