feat: extend batch processing and enhanced progress to CSV and PDF imports

Applies the same performance optimizations from v3.38.0 Excel improvements
to CSV and PDF importers:

- CSV: Batch processing with 10 rows per chunk
- PDF: Batch processing with 5 sections per chunk
- Both: Parallel entity + concept extraction
- Both: Enhanced progress reporting with throughput and ETA
- Performance tests for both formats

Performance results:
- CSV: 9,091 rows/sec with 93.8% cache hit rate
- PDF: 313 sections/sec with 90.2% cache hit rate

All formats now have consistent batch processing architecture
and real-time progress feedback.
This commit is contained in:
David Snelling 2025-10-13 10:32:25 -07:00
parent 77c104a9a4
commit bb46da2ee7
4 changed files with 466 additions and 135 deletions

View file

@ -41,12 +41,18 @@ export interface SmartCSVOptions extends FormatHandlerOptions {
csvDelimiter?: string
csvHeaders?: boolean
/** Progress callback */
/** Progress callback (v3.39.0: Enhanced with performance metrics) */
onProgress?: (stats: {
processed: number
total: number
entities: number
relationships: number
/** Rows per second (v3.39.0) */
throughput?: number
/** Estimated time remaining in ms (v3.39.0) */
eta?: number
/** Current phase (v3.39.0) */
phase?: string
}) => void
}
@ -174,7 +180,7 @@ export class SmartCSVImporter {
// Detect column names
const columns = this.detectColumns(rows[0], opts)
// Process each row
// Process each row with BATCHED PARALLEL PROCESSING (v3.39.0)
const extractedRows: ExtractedRow[] = []
const entityMap = new Map<string, string>()
const stats = {
@ -182,126 +188,161 @@ export class SmartCSVImporter {
byConfidence: { high: 0, medium: 0, low: 0 }
}
for (let i = 0; i < rows.length; i++) {
const row = rows[i]
// Batch processing configuration
const CHUNK_SIZE = 10 // Process 10 rows at a time for optimal performance
let totalProcessed = 0
const performanceStartTime = Date.now()
// Extract data from row
const term = this.getColumnValue(row, columns.term) || `Entity_${i}`
const definition = this.getColumnValue(row, columns.definition) || ''
const type = this.getColumnValue(row, columns.type)
const relatedTerms = this.getColumnValue(row, columns.related)
// Process rows in chunks
for (let chunkStart = 0; chunkStart < rows.length; chunkStart += CHUNK_SIZE) {
const chunk = rows.slice(chunkStart, Math.min(chunkStart + CHUNK_SIZE, rows.length))
// Extract entities from definition
let relatedEntities: ExtractedEntity[] = []
if (opts.enableNeuralExtraction && definition) {
relatedEntities = await this.extractor.extract(definition, {
confidence: opts.confidenceThreshold * 0.8, // Lower threshold for related entities
neuralMatching: true,
cache: { enabled: true }
})
// Process chunk in parallel for massive speedup
const chunkResults = await Promise.all(
chunk.map(async (row, chunkIndex) => {
const i = chunkStart + chunkIndex
// Filter out the main term from related entities
relatedEntities = relatedEntities.filter(
e => e.text.toLowerCase() !== term.toLowerCase()
)
}
// Extract data from row
const term = this.getColumnValue(row, columns.term) || `Entity_${i}`
const definition = this.getColumnValue(row, columns.definition) || ''
const type = this.getColumnValue(row, columns.type)
const relatedTerms = this.getColumnValue(row, columns.related)
// Determine main entity type
const mainEntityType = type ?
this.mapTypeString(type) :
(relatedEntities.length > 0 ? relatedEntities[0].type : NounType.Thing)
// Parallel extraction: entities AND concepts at the same time
const [relatedEntities, concepts] = await Promise.all([
// Extract entities from definition
opts.enableNeuralExtraction && definition
? this.extractor.extract(definition, {
confidence: opts.confidenceThreshold * 0.8,
neuralMatching: true,
cache: { enabled: true }
}).then(entities =>
// Filter out the main term from related entities
entities.filter(e => e.text.toLowerCase() !== term.toLowerCase())
)
: Promise.resolve([]),
// Generate entity ID
const entityId = this.generateEntityId(term)
entityMap.set(term.toLowerCase(), entityId)
// Extract concepts (in parallel with entity extraction)
opts.enableConceptExtraction && definition
? this.brain.extractConcepts(definition, { limit: 10 }).catch(() => [])
: Promise.resolve([])
])
// Extract concepts
let concepts: string[] = []
if (opts.enableConceptExtraction && definition) {
try {
concepts = await this.brain.extractConcepts(definition, { limit: 10 })
} catch (error) {
concepts = []
}
}
// Determine main entity type
const mainEntityType = type ?
this.mapTypeString(type) :
(relatedEntities.length > 0 ? relatedEntities[0].type : NounType.Thing)
// Create main entity
const mainEntity = {
id: entityId,
name: term,
type: mainEntityType,
description: definition,
confidence: 0.95, // Main entity from row has high confidence
metadata: {
source: 'csv',
row: i + 1,
originalData: row,
concepts,
extractedAt: Date.now()
}
}
// Generate entity ID
const entityId = this.generateEntityId(term)
// Track statistics
this.updateStats(stats, mainEntityType, mainEntity.confidence)
// Infer relationships
const relationships: ExtractedRow['relationships'] = []
if (opts.enableRelationshipInference) {
// Extract relationships from definition text
for (const relEntity of relatedEntities) {
const verbType = await this.inferRelationship(
term,
relEntity.text,
definition
)
relationships.push({
from: entityId,
to: relEntity.text,
type: verbType,
confidence: relEntity.confidence,
evidence: `Extracted from: "${definition.substring(0, 100)}..."`
})
}
// Parse explicit "Related" column
if (relatedTerms) {
const terms = relatedTerms.split(/[,;|]/).map(t => t.trim()).filter(Boolean)
for (const relTerm of terms) {
// Ensure we don't create self-relationships
if (relTerm.toLowerCase() !== term.toLowerCase()) {
relationships.push({
from: entityId,
to: relTerm,
type: VerbType.RelatedTo,
confidence: 0.9,
evidence: `Explicitly listed in "Related" column`
})
// Create main entity
const mainEntity = {
id: entityId,
name: term,
type: mainEntityType,
description: definition,
confidence: 0.95,
metadata: {
source: 'csv',
row: i + 1,
originalData: row,
concepts,
extractedAt: Date.now()
}
}
}
// Infer relationships
const relationships: ExtractedRow['relationships'] = []
if (opts.enableRelationshipInference) {
// Extract relationships from definition text
for (const relEntity of relatedEntities) {
const verbType = await this.inferRelationship(
term,
relEntity.text,
definition
)
relationships.push({
from: entityId,
to: relEntity.text,
type: verbType,
confidence: relEntity.confidence,
evidence: `Extracted from: "${definition.substring(0, 100)}..."`
})
}
// Parse explicit "Related" column
if (relatedTerms) {
const terms = relatedTerms.split(/[,;|]/).map(t => t.trim()).filter(Boolean)
for (const relTerm of terms) {
if (relTerm.toLowerCase() !== term.toLowerCase()) {
relationships.push({
from: entityId,
to: relTerm,
type: VerbType.RelatedTo,
confidence: 0.9,
evidence: `Explicitly listed in "Related" column`
})
}
}
}
}
return {
term,
entityId,
mainEntity,
mainEntityType,
relatedEntities,
relationships,
concepts
}
})
)
// Process chunk results sequentially to maintain order
for (const result of chunkResults) {
// Store entity ID mapping
entityMap.set(result.term.toLowerCase(), result.entityId)
// Track statistics
this.updateStats(stats, result.mainEntityType, result.mainEntity.confidence)
// Add extracted row
extractedRows.push({
entity: result.mainEntity,
relatedEntities: result.relatedEntities.map(e => ({
name: e.text,
type: e.type,
confidence: e.confidence
})),
relationships: result.relationships,
concepts: result.concepts
})
}
// Add extracted row
extractedRows.push({
entity: mainEntity,
relatedEntities: relatedEntities.map(e => ({
name: e.text,
type: e.type,
confidence: e.confidence
})),
relationships,
concepts
})
// Update progress tracking
totalProcessed += chunk.length
// Report progress
// Calculate performance metrics
const elapsed = Date.now() - performanceStartTime
const rowsPerSecond = totalProcessed / (elapsed / 1000)
const remainingRows = rows.length - totalProcessed
const estimatedTimeRemaining = remainingRows / rowsPerSecond
// Report progress with enhanced metrics
opts.onProgress({
processed: i + 1,
processed: totalProcessed,
total: rows.length,
entities: extractedRows.length + relatedEntities.length,
relationships: relationships.length
})
entities: extractedRows.reduce((sum, row) => sum + 1 + row.relatedEntities.length, 0),
relationships: extractedRows.reduce((sum, row) => sum + row.relationships.length, 0),
// Additional performance metrics (v3.39.0)
throughput: Math.round(rowsPerSecond * 10) / 10,
eta: Math.round(estimatedTimeRemaining),
phase: 'extracting'
} as any)
}
return {

View file

@ -39,12 +39,18 @@ export interface SmartPDFOptions extends FormatHandlerOptions {
/** Group by page or full document */
groupBy?: 'page' | 'document'
/** Progress callback */
/** Progress callback (v3.39.0: Enhanced with performance metrics) */
onProgress?: (stats: {
processed: number
total: number
entities: number
relationships: number
/** Sections per second (v3.39.0) */
throughput?: number
/** Estimated time remaining in ms (v3.39.0) */
eta?: number
/** Current phase (v3.39.0) */
phase?: string
}) => void
}
@ -183,7 +189,7 @@ export class SmartPDFImporter {
// Group data by page or combine into single document
const grouped = this.groupData(data, opts)
// Process each group
// Process each group with BATCHED PARALLEL PROCESSING (v3.39.0)
const sections: ExtractedSection[] = []
const entityMap = new Map<string, string>()
const stats = {
@ -192,20 +198,44 @@ export class SmartPDFImporter {
bySource: { paragraphs: 0, tables: 0 }
}
let processedCount = 0
// Batch processing configuration
const CHUNK_SIZE = 5 // Process 5 sections at a time (smaller than rows due to section size)
let totalProcessed = 0
const performanceStartTime = Date.now()
const totalGroups = grouped.length
for (const group of grouped) {
const sectionResult = await this.processSection(group, opts, stats, entityMap)
sections.push(sectionResult)
// Process sections in chunks
for (let chunkStart = 0; chunkStart < grouped.length; chunkStart += CHUNK_SIZE) {
const chunk = grouped.slice(chunkStart, Math.min(chunkStart + CHUNK_SIZE, grouped.length))
processedCount++
// Process chunk in parallel for better performance
const chunkResults = await Promise.all(
chunk.map(group => this.processSection(group, opts, stats, entityMap))
)
// Add results sequentially to maintain order
sections.push(...chunkResults)
// Update progress tracking
totalProcessed += chunk.length
// Calculate performance metrics
const elapsed = Date.now() - performanceStartTime
const sectionsPerSecond = totalProcessed / (elapsed / 1000)
const remainingSections = grouped.length - totalProcessed
const estimatedTimeRemaining = remainingSections / sectionsPerSecond
// Report progress with enhanced metrics
opts.onProgress({
processed: processedCount,
processed: totalProcessed,
total: totalGroups,
entities: sections.reduce((sum, s) => sum + s.entities.length, 0),
relationships: sections.reduce((sum, s) => sum + s.relationships.length, 0)
})
relationships: sections.reduce((sum, s) => sum + s.relationships.length, 0),
// Additional performance metrics (v3.39.0)
throughput: Math.round(sectionsPerSecond * 10) / 10,
eta: Math.round(estimatedTimeRemaining),
phase: 'extracting'
} as any)
}
const pagesProcessed = new Set(data.map(d => d._page)).size
@ -298,25 +328,22 @@ export class SmartPDFImporter {
const combinedText = texts.join('\n\n')
// Extract entities if enabled
let extractedEntities: ExtractedEntity[] = []
if (options.enableNeuralExtraction && combinedText.length > 0) {
extractedEntities = await this.extractor.extract(combinedText, {
confidence: options.confidenceThreshold || 0.6,
neuralMatching: true,
cache: { enabled: true }
})
}
// Parallel extraction: entities AND concepts at the same time (v3.39.0)
const [extractedEntities, concepts] = await Promise.all([
// Extract entities if enabled
options.enableNeuralExtraction && combinedText.length > 0
? this.extractor.extract(combinedText, {
confidence: options.confidenceThreshold || 0.6,
neuralMatching: true,
cache: { enabled: true }
})
: Promise.resolve([]),
// Extract concepts if enabled
let concepts: string[] = []
if (options.enableConceptExtraction && combinedText.length > 0) {
try {
concepts = await this.brain.extractConcepts(combinedText, { limit: 15 })
} catch (error) {
concepts = []
}
}
// Extract concepts (in parallel with entity extraction)
options.enableConceptExtraction && combinedText.length > 0
? this.brain.extractConcepts(combinedText, { limit: 15 }).catch(() => [])
: Promise.resolve([])
])
// Create entity objects
const entities = extractedEntities.map(e => {