File size: 9,984 Bytes
d520909 |
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 |
"""
Extraction Schemas for Document Intelligence
Pydantic models for schema-based field extraction, tables, and charts.
"""
from enum import Enum
from typing import List, Dict, Any, Optional, Union
from pydantic import BaseModel, Field
from .core import BoundingBox, EvidenceRef
class FieldType(str, Enum):
"""Supported field types for extraction."""
STRING = "string"
INTEGER = "integer"
FLOAT = "float"
BOOLEAN = "boolean"
DATE = "date"
CURRENCY = "currency"
PERCENTAGE = "percentage"
EMAIL = "email"
PHONE = "phone"
ADDRESS = "address"
LIST = "list"
OBJECT = "object"
class FieldDefinition(BaseModel):
"""
Definition of a field to extract from a document.
Used to build extraction schemas.
"""
name: str = Field(..., description="Field name/key")
type: FieldType = Field(..., description="Expected data type")
description: str = Field(..., description="Human-readable description")
required: bool = Field(default=False, description="Whether field is required")
# Validation constraints
pattern: Optional[str] = Field(default=None, description="Regex pattern for validation")
min_value: Optional[float] = Field(default=None, description="Minimum numeric value")
max_value: Optional[float] = Field(default=None, description="Maximum numeric value")
enum_values: Optional[List[str]] = Field(default=None, description="Allowed values")
# Extraction hints
aliases: List[str] = Field(
default_factory=list,
description="Alternative names/labels for the field"
)
search_context: Optional[str] = Field(
default=None,
description="Context hint for where to find this field"
)
# Nested fields (for object/list types)
nested_fields: Optional[List["FieldDefinition"]] = Field(
default=None,
description="Nested field definitions for complex types"
)
class ExtractionSchema(BaseModel):
"""
Schema defining fields to extract from a document.
Supports document-type-specific extraction rules.
"""
schema_id: str = Field(..., description="Unique schema identifier")
name: str = Field(..., description="Human-readable schema name")
description: str = Field(..., description="Schema description")
version: str = Field(default="1.0", description="Schema version")
# Field definitions
fields: List[FieldDefinition] = Field(
default_factory=list,
description="Fields to extract"
)
# Document type association
document_types: List[str] = Field(
default_factory=list,
description="Applicable document types"
)
# Validation rules
cross_field_validations: List[str] = Field(
default_factory=list,
description="Cross-field validation expressions"
)
# Extraction configuration
require_evidence: bool = Field(
default=True,
description="Require evidence for all extracted fields"
)
min_confidence: float = Field(
default=0.7,
ge=0.0,
le=1.0,
description="Minimum confidence threshold"
)
abstain_on_low_confidence: bool = Field(
default=True,
description="Abstain rather than guess when confidence is low"
)
def get_field(self, name: str) -> Optional[FieldDefinition]:
"""Get field definition by name."""
for field in self.fields:
if field.name == name or name in field.aliases:
return field
return None
def get_required_fields(self) -> List[FieldDefinition]:
"""Get all required field definitions."""
return [f for f in self.fields if f.required]
class TableCell(BaseModel):
"""
Single cell in a table structure.
"""
cell_id: str = Field(..., description="Unique cell identifier")
row: int = Field(..., ge=0, description="Row index (0-based)")
col: int = Field(..., ge=0, description="Column index (0-based)")
text: str = Field(..., description="Cell text content")
bbox: BoundingBox = Field(..., description="Cell bounding box")
# Span information
row_span: int = Field(default=1, ge=1, description="Number of rows spanned")
col_span: int = Field(default=1, ge=1, description="Number of columns spanned")
# Cell type
is_header: bool = Field(default=False, description="Whether cell is a header")
is_empty: bool = Field(default=False, description="Whether cell is empty")
# Confidence
confidence: float = Field(default=1.0, ge=0.0, le=1.0)
class TableData(BaseModel):
"""
Structured table data extracted from a document.
"""
table_id: str = Field(..., description="Unique table identifier")
page: int = Field(..., ge=0, description="Page number")
bbox: BoundingBox = Field(..., description="Table bounding box")
# Structure
num_rows: int = Field(..., ge=1, description="Number of rows")
num_cols: int = Field(..., ge=1, description="Number of columns")
cells: List[TableCell] = Field(default_factory=list, description="All cells")
# Headers
header_rows: List[int] = Field(
default_factory=list,
description="Row indices that are headers"
)
header_cols: List[int] = Field(
default_factory=list,
description="Column indices that are headers"
)
# Caption
caption: Optional[str] = Field(default=None, description="Table caption")
caption_bbox: Optional[BoundingBox] = Field(default=None)
# Confidence
confidence: float = Field(default=1.0, ge=0.0, le=1.0)
# Evidence
evidence: Optional[EvidenceRef] = Field(default=None)
def to_markdown(self) -> str:
"""Convert table to markdown format."""
if not self.cells:
return ""
# Build grid
grid = [[None for _ in range(self.num_cols)] for _ in range(self.num_rows)]
for cell in self.cells:
if cell.row < self.num_rows and cell.col < self.num_cols:
grid[cell.row][cell.col] = cell.text
# Generate markdown
lines = []
for i, row in enumerate(grid):
line = "| " + " | ".join(str(c) if c else "" for c in row) + " |"
lines.append(line)
if i == 0 or i in self.header_rows:
lines.append("|" + "|".join(["---"] * self.num_cols) + "|")
return "\n".join(lines)
def to_dict_list(self) -> List[Dict[str, str]]:
"""Convert table to list of dictionaries (using first row as keys)."""
if not self.cells or self.num_rows < 2:
return []
# Build grid
grid = [[None for _ in range(self.num_cols)] for _ in range(self.num_rows)]
for cell in self.cells:
if cell.row < self.num_rows and cell.col < self.num_cols:
grid[cell.row][cell.col] = cell.text
# Use first row as headers
headers = [str(h) if h else f"col_{i}" for i, h in enumerate(grid[0])]
# Build list of dicts
result = []
for row in grid[1:]:
row_dict = {headers[i]: str(v) if v else "" for i, v in enumerate(row)}
result.append(row_dict)
return result
class ChartType(str, Enum):
"""Types of charts/graphs."""
BAR = "bar"
LINE = "line"
PIE = "pie"
SCATTER = "scatter"
AREA = "area"
HISTOGRAM = "histogram"
BOX = "box"
HEATMAP = "heatmap"
TREEMAP = "treemap"
FLOWCHART = "flowchart"
DIAGRAM = "diagram"
OTHER = "other"
class ChartData(BaseModel):
"""
Structured chart/graph data extracted from a document.
"""
chart_id: str = Field(..., description="Unique chart identifier")
page: int = Field(..., ge=0, description="Page number")
bbox: BoundingBox = Field(..., description="Chart bounding box")
chart_type: ChartType = Field(..., description="Type of chart")
# Chart content
title: Optional[str] = Field(default=None, description="Chart title")
x_axis_label: Optional[str] = Field(default=None, description="X-axis label")
y_axis_label: Optional[str] = Field(default=None, description="Y-axis label")
# Data series
series: List[Dict[str, Any]] = Field(
default_factory=list,
description="Data series extracted from chart"
)
# Trends and insights
trends: List[str] = Field(
default_factory=list,
description="Identified trends or patterns"
)
# Caption
caption: Optional[str] = Field(default=None, description="Chart caption")
# Confidence and evidence
confidence: float = Field(default=1.0, ge=0.0, le=1.0)
evidence: Optional[EvidenceRef] = Field(default=None)
# Raw description (for LLM extraction)
description: Optional[str] = Field(
default=None,
description="Natural language description of the chart"
)
class ExtractedField(BaseModel):
"""
A single extracted field value with evidence.
"""
field_name: str = Field(..., description="Field name from schema")
value: Any = Field(..., description="Extracted value")
confidence: float = Field(..., ge=0.0, le=1.0, description="Extraction confidence")
evidence: List[EvidenceRef] = Field(
default_factory=list,
description="Evidence supporting the extraction"
)
# Validation status
is_valid: bool = Field(default=True, description="Whether value passed validation")
validation_errors: List[str] = Field(
default_factory=list,
description="Validation error messages"
)
# Abstention
abstained: bool = Field(
default=False,
description="Whether extraction was abstained"
)
abstain_reason: Optional[str] = Field(
default=None,
description="Reason for abstention"
)
@property
def is_grounded(self) -> bool:
"""Check if extraction has evidence."""
return len(self.evidence) > 0 and not self.abstained
|