feat(labelling): add primitive labelling algorithm

Implement an algorithm that is a vast simplification of CRF. It simply iterates over the sentence once, starting from the beginning, and chooses the best label for the word by evaluating each feature function.

Author: chriscasola

labelling/labelling.go

@@ -0,0 +1,40 @@
+package labelling
+
+// FeatureFunction is a feature function for the model
+type FeatureFunction func(sentence []string, i int, labelCurr string, labelPrev string) float64
+
+// Feature includes the weight and feature function for a feature
+type Feature struct {
+Weight float64
+Value FeatureFunction
+}
+
+// FindBestLabelling determines the best labeling for the given sentence
+func FindBestLabelling(sentence []string, labels []string, features []Feature) []string {
+labelling := make([]string, 0)
+
+for i := 0; i < len(sentence); i++ {
+bestScore, bestLabel, currentScore := -1.0, "", 0.0
+prevLabel := ""
+if i > 0 {
+prevLabel = labelling[i-1]
+}
+
+for j := 0; j < len(labels); j++ {
+for k := 0; k < len(features); k++ {
+currentScore += (features[k].Weight * features[k].Value(sentence, i, labels[j], prevLabel))
+}
+
+if currentScore > bestScore {
+bestScore = currentScore
+bestLabel = labels[j]
+}
+
+currentScore = 0
+}
+
+labelling = append(labelling, bestLabel)
+}
+
+return labelling
+}

labelling/labelling_test.go

@@ -0,0 +1,58 @@
+package labelling
+
+import (
+"reflect"
+"strings"
+"testing"
+)
+
+func stringInArray(list []string, s string) bool {
+for i := 0; i < len(list); i++ {
+if strings.ToLower(list[i]) == strings.ToLower(s) {
+return true
+}
+}
+
+return false
+}
+
+func isQuantityAtBeginning(sentence []string, i int, labelCurr string, labelPrev string) float64 {
+if i == 0 && stringInArray([]string{"1", "2", "3", "4", "5", "6", "7", "8", "9"}, sentence[i]) && labelCurr == "quantity" {
+return 1
+}
+
+return 0
+}
+
+func unitFollowsQuantity(sentence []string, i int, labelCurr string, labelPrev string) float64 {
+if labelPrev == "quantity" && labelCurr == "units" {
+return 1
+}
+
+return 0
+}
+
+func ingredientFollowsUnit(sentence []string, i int, labelCurr string, labelPrev string) float64 {
+if labelPrev == "units" && labelCurr == "ingredient" {
+return 1
+}
+
+return 0
+}
+
+func TestFindBestLabelling(t *testing.T) {
+sentence := strings.Split("1 cup apples", " ")
+labels := []string{"quantity", "units", "ingredient"}
+features := []Feature{
+{1.0, isQuantityAtBeginning},
+{1.0, unitFollowsQuantity},
+{1.0, ingredientFollowsUnit},
+}
+bestLabelling := FindBestLabelling(sentence, labels, features)
+
+expectedLabelling := []string{"quantity", "units", "ingredient"}
+
+if reflect.DeepEqual(bestLabelling, expectedLabelling) != true {
+t.Errorf("Expected %v to equal %v", bestLabelling, expectedLabelling)
+}
+}