fluxrec: personal Miniflux article recommender

1 files changed, 76 insertions, 0 deletions

diff --git a/core/ml_test.go b/core/ml_test.go
new file mode 100644
index 0000000..3623b39
--- /dev/null
+++ b/core/ml_test.go
@@ -0,0 +1,76 @@
+// Roundtrip test for the vendored TF-IDF + LR pipeline: fit a tiny model,
+// confirm scores separate the two synthetic topics. Guards against a broken
+// copy more than against algorithm regressions.
+package core
+
+import (
+	"testing"
+)
+
+func TestTFIDFLRRoundtrip(t *testing.T) {
+	// Two separable synthetic "topics": optics vs cooking.
+	docs := []string{
+		"quantum laser optics photon",
+		"laser photon quantum measurement",
+		"photon optics quantum laser",
+		"optics quantum laser photon",
+		"recipe cooking knife onion",
+		"onion recipe cooking garlic",
+		"cooking garlic onion skillet",
+		"garlic recipe onion cooking",
+	}
+	labels := []float64{1, 1, 1, 1, 0, 0, 0, 0}
+
+	v := &TFIDFVectorizer{NgramMin: 1, NgramMax: 2, MinDF: 1, MaxDF: 0.9}
+	v.Fit(docs)
+	if len(v.OrderedVocab) == 0 {
+		t.Fatal("vectorizer learned empty vocabulary")
+	}
+
+	vecs := v.Transform(docs)
+	lr := (&LogisticRegression{Lambda: 0.001}).Validate()
+	weights, err := lr.Fit(vecs, labels, nil)
+	if err != nil {
+		t.Fatalf("Fit: %v", err)
+	}
+
+	// Persist and reload through the envelope, as serve does.
+	model := &ModelEnvelope{
+		Vectorizer:   v.Vocabulary,
+		OrderedVocab: v.OrderedVocab,
+		Weights:      weights,
+	}
+	rv := CreateVectorizerFromModel(model)
+
+	opticsScore := mustScore(t, rv, model, "quantum photon laser experiment")
+	cookingScore := mustScore(t, rv, model, "onion garlic skillet dinner")
+	if !(opticsScore > 0.5 && cookingScore < 0.5) {
+		t.Fatalf("expected separation, got optics=%.3f cooking=%.3f", opticsScore, cookingScore)
+	}
+	if opticsScore <= cookingScore {
+		t.Fatalf("expected optics score above cooking score, got %.3f <= %.3f", opticsScore, cookingScore)
+	}
+}
+
+func mustScore(t *testing.T, v *TFIDFVectorizer, m *ModelEnvelope, title string) float64 {
+	t.Helper()
+	vec := v.Transform([]string{title})
+	s, err := PredictScore(vec[0], m.Weights)
+	if err != nil {
+		t.Fatalf("PredictScore(%q): %v", title, err)
+	}
+	return s
+}
+
+func TestTokenize(t *testing.T) {
+	got := Tokenize("The Fall of the House of Usher!")
+	want := []string{"fall", "house", "usher"}
+	if len(got) != len(want) {
+		t.Fatalf("Tokenize: got %v, want %v", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("Tokenize: got %v, want %v", got, want)
+		}
+	}
+}