fix: Update production API base URL in documentation and configuration files. Correct the URL in the desktop app README, migration documentation, and Vite configuration. Enhance ONNX model feature handling in the processing pipeline to align with the updated feature set, ensuring compatibility with the latest model version. Add tests for feature extraction to validate changes.
This commit is contained in:
erik
@@ -7,15 +7,40 @@ import (
|
||||
"unicode"
|
||||
)
|
||||
|
||||
// FeatureCount debe coincidir con models v1.1.0 (services-count-facturas-medicas-ml).
|
||||
const FeatureCount = 14
|
||||
|
||||
var (
|
||||
codePattern = regexp.MustCompile(`^\d{4,6}$`)
|
||||
currencyPattern = regexp.MustCompile(`^\d{1,3}(,\d{3})+(\.\d+)?$`)
|
||||
codePattern = regexp.MustCompile(`^\d{4,6}$`)
|
||||
serviceCodePattern = regexp.MustCompile(`(?i)^(\d{4,8}|\d+-\d+|[A-Z]{2,}\d{2,})$`)
|
||||
currencyPattern = regexp.MustCompile(`^\d{1,3}(,\d{3})+(\.\d+)?$`)
|
||||
headerKeywordPattern = regexp.MustCompile(`(?i)\b(FECHA|VALOR|SUBTOTAL|CUFE|RESOLUCION|CONTRATO|FACTURA|NIT|TELEFONO|TOTAL)\b`)
|
||||
)
|
||||
|
||||
type rowFeatures struct {
|
||||
Features [8]float32
|
||||
TextoFila string
|
||||
Cluster int
|
||||
Features [FeatureCount]float32
|
||||
TextoFila string
|
||||
Cluster int
|
||||
yNorm float32
|
||||
// índices usados en el paso de continuación
|
||||
hasServiceTablePattern int
|
||||
currencyCount int
|
||||
}
|
||||
|
||||
func isServiceCode(token string) bool {
|
||||
return serviceCodePattern.MatchString(token)
|
||||
}
|
||||
|
||||
func hasHeaderKeyword(text string) bool {
|
||||
return headerKeywordPattern.MatchString(text)
|
||||
}
|
||||
|
||||
func firstTokenOf(text string) string {
|
||||
parts := strings.Fields(text)
|
||||
if len(parts) == 0 {
|
||||
return ""
|
||||
}
|
||||
return parts[0]
|
||||
}
|
||||
|
||||
func extractRowFeatures(group []indexedBox, pageW, pageH int) rowFeatures {
|
||||
@@ -54,13 +79,29 @@ func extractRowFeatures(group []indexedBox, pageW, pageH int) rowFeatures {
|
||||
ySum += b.Y
|
||||
}
|
||||
|
||||
xSpanNorm := 0.0
|
||||
xSpanNorm := float32(0)
|
||||
if pageW > 0 {
|
||||
xSpanNorm = float64(xMax-xMin) / float64(pageW)
|
||||
xSpanNorm = float32(xMax-xMin) / float32(pageW)
|
||||
}
|
||||
yNorm := 0.0
|
||||
yNorm := float32(0)
|
||||
if pageH > 0 {
|
||||
yNorm = float64(ySum/len(group)) / float64(pageH)
|
||||
yNorm = float32(ySum/len(group)) / float32(pageH)
|
||||
}
|
||||
firstTokenXNorm := float32(0)
|
||||
if pageW > 0 {
|
||||
firstTokenXNorm = float32(xMin) / float32(pageW)
|
||||
}
|
||||
|
||||
tailStart := 0
|
||||
if len(texts) >= 3 {
|
||||
tailStart = len(texts) - 3
|
||||
}
|
||||
currencyAtRowEnd := 0
|
||||
for _, t := range texts[tailStart:] {
|
||||
if currencyPattern.MatchString(t) {
|
||||
currencyAtRowEnd = 1
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
totalChars := len(textoFila)
|
||||
@@ -83,22 +124,73 @@ func extractRowFeatures(group []indexedBox, pageW, pageH int) rowFeatures {
|
||||
firstIsCode = 1
|
||||
}
|
||||
|
||||
hasTablePattern := 0
|
||||
if currencyCount >= 2 && integerCount >= 1 {
|
||||
hasTablePattern = 1
|
||||
}
|
||||
headerKW := 0
|
||||
if hasHeaderKeyword(textoFila) {
|
||||
headerKW = 1
|
||||
}
|
||||
|
||||
return rowFeatures{
|
||||
Features: [8]float32{
|
||||
Features: [FeatureCount]float32{
|
||||
float32(firstIsCode),
|
||||
boolToFloat(isServiceCode(firstToken)),
|
||||
float32(hasTablePattern),
|
||||
float32(headerKW),
|
||||
float32(currencyAtRowEnd),
|
||||
firstTokenXNorm,
|
||||
0, // is_likely_continuation — segundo paso
|
||||
float32(currencyCount),
|
||||
float32(integerCount),
|
||||
float32(xSpanNorm),
|
||||
float32(yNorm),
|
||||
xSpanNorm,
|
||||
yNorm,
|
||||
float32(len(texts)),
|
||||
float32(special) / float32(totalChars),
|
||||
float32(digits) / float32(totalChars),
|
||||
},
|
||||
TextoFila: textoFila,
|
||||
Cluster: group[0].Cluster,
|
||||
TextoFila: textoFila,
|
||||
Cluster: group[0].Cluster,
|
||||
yNorm: yNorm,
|
||||
hasServiceTablePattern: hasTablePattern,
|
||||
currencyCount: currencyCount,
|
||||
}
|
||||
}
|
||||
|
||||
func boolToFloat(v bool) float32 {
|
||||
if v {
|
||||
return 1
|
||||
}
|
||||
return 0
|
||||
}
|
||||
|
||||
func buildFeatureRows(boxes []indexedBox, pageW, pageH int) []rowFeatures {
|
||||
groups := groupByCluster(boxes)
|
||||
rows := make([]rowFeatures, 0, len(groups))
|
||||
for _, group := range groups {
|
||||
rows = append(rows, extractRowFeatures(group, pageW, pageH))
|
||||
}
|
||||
if len(rows) == 0 {
|
||||
return rows
|
||||
}
|
||||
|
||||
sort.Slice(rows, func(i, j int) bool { return rows[i].yNorm < rows[j].yNorm })
|
||||
|
||||
for i := 1; i < len(rows); i++ {
|
||||
prev := rows[i-1]
|
||||
cur := &rows[i]
|
||||
firstToken := firstTokenOf(cur.TextoFila)
|
||||
if prev.hasServiceTablePattern == 1 &&
|
||||
cur.currencyCount == 0 &&
|
||||
!isServiceCode(firstToken) &&
|
||||
!codePattern.MatchString(firstToken) {
|
||||
cur.Features[6] = 1
|
||||
}
|
||||
}
|
||||
return rows
|
||||
}
|
||||
|
||||
func isDigits(s string) bool {
|
||||
if s == "" {
|
||||
return false
|
||||
@@ -110,13 +202,3 @@ func isDigits(s string) bool {
|
||||
}
|
||||
return true
|
||||
}
|
||||
|
||||
func buildFeatureRows(boxes []indexedBox, pageW, pageH int) []rowFeatures {
|
||||
groups := groupByCluster(boxes)
|
||||
order := sortClusters(groups)
|
||||
rows := make([]rowFeatures, 0, len(groups))
|
||||
for _, cid := range order {
|
||||
rows = append(rows, extractRowFeatures(groups[cid], pageW, pageH))
|
||||
}
|
||||
return rows
|
||||
}
|
||||
|
||||
@@ -0,0 +1,48 @@
|
||||
package pipeline
|
||||
|
||||
import "testing"
|
||||
|
||||
func TestExtractRowFeaturesServiceTablePattern(t *testing.T) {
|
||||
group := []indexedBox{
|
||||
{Text: "890701", X: 10, Y: 200, Cluster: 0},
|
||||
{Text: "CONSULTA", X: 100, Y: 201, Cluster: 0},
|
||||
{Text: "16,277.00", X: 400, Y: 200, Cluster: 0},
|
||||
{Text: "1", X: 500, Y: 201, Cluster: 0},
|
||||
{Text: "16,277.00", X: 600, Y: 200, Cluster: 0},
|
||||
}
|
||||
row := extractRowFeatures(group, 1000, 2000)
|
||||
if row.Features[2] != 1 {
|
||||
t.Fatalf("has_service_table_pattern = %v, want 1", row.Features[2])
|
||||
}
|
||||
if row.Features[4] != 1 {
|
||||
t.Fatalf("currency_at_row_end = %v, want 1", row.Features[4])
|
||||
}
|
||||
}
|
||||
|
||||
func TestBuildFeaturesContinuation(t *testing.T) {
|
||||
boxes := []indexedBox{
|
||||
{Text: "890701", X: 10, Y: 100, Cluster: 0},
|
||||
{Text: "CONSULTA", X: 100, Y: 100, Cluster: 0},
|
||||
{Text: "16,277.00", X: 400, Y: 100, Cluster: 0},
|
||||
{Text: "1", X: 500, Y: 100, Cluster: 0},
|
||||
{Text: "16,277.00", X: 600, Y: 100, Cluster: 0},
|
||||
{Text: "DE", X: 100, Y: 120, Cluster: 1},
|
||||
{Text: "SEGUIMIENTO", X: 200, Y: 120, Cluster: 1},
|
||||
}
|
||||
rows := buildFeatureRows(boxes, 1000, 2000)
|
||||
if len(rows) != 2 {
|
||||
t.Fatalf("rows = %d, want 2", len(rows))
|
||||
}
|
||||
if rows[0].Features[2] != 1 {
|
||||
t.Fatal("first row should have service table pattern")
|
||||
}
|
||||
if rows[1].Features[6] != 1 {
|
||||
t.Fatal("second row should be likely continuation")
|
||||
}
|
||||
}
|
||||
|
||||
func TestFeatureCount(t *testing.T) {
|
||||
if FeatureCount != 14 {
|
||||
t.Fatalf("FeatureCount = %d, want 14", FeatureCount)
|
||||
}
|
||||
}
|
||||
@@ -88,9 +88,9 @@ func (p *Processor) ProcessPDF(filename string, pdfData []byte) domain.PDFResult
|
||||
return result
|
||||
}
|
||||
|
||||
features := make([][8]float32, len(rows))
|
||||
features := make([][]float32, len(rows))
|
||||
for i, r := range rows {
|
||||
features[i] = r.Features
|
||||
features[i] = r.Features[:]
|
||||
}
|
||||
|
||||
tInfer := time.Now()
|
||||
|
||||
Reference in New Issue
Block a user