fluxrec: personal Miniflux article recommender

1 files changed, 103 insertions, 0 deletions

diff --git a/cmds/score.go b/cmds/score.go
new file mode 100644
index 0000000..da6167a
--- /dev/null
+++ b/cmds/score.go
@@ -0,0 +1,103 @@
+// Score command: debugging utility, one trained model vs. stdin titles.
+//
+// Reads one title per line on stdin, writes {"title", "score"} JSONL on
+// stdout. Exists to make the vendored ML core exercisable end-to-end before
+// the real commands (export/train/serve) land, and stays useful later for
+// poking at model behavior by hand.
+package cmds
+
+import (
+	"bufio"
+	"encoding/json"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"fluxrec/core"
+)
+
+type ScoreCommand struct {
+	ModelPath string
+}
+
+func (c *ScoreCommand) Name() string { return "score" }
+
+func (c *ScoreCommand) Init(args []string) error {
+	fs := flag.NewFlagSet(c.Name(), flag.ContinueOnError)
+	fs.Usage = func() {
+		fmt.Fprint(fs.Output(), `Usage: fluxrec score --model MODEL
+
+Score stdin titles (one per line) with a trained model, JSONL to stdout.
+`)
+		fs.PrintDefaults()
+	}
+	fs.StringVar(&c.ModelPath, "model", "", "Path to model.json (required)")
+	if err := fs.Parse(args); err != nil {
+		return err
+	}
+	if c.ModelPath == "" {
+		return errors.New("--model flag is required")
+	}
+	clean := filepath.Clean(c.ModelPath)
+	if strings.Contains(clean, "..") {
+		return fmt.Errorf("invalid model path %q: must not contain '..'", c.ModelPath)
+	}
+	c.ModelPath = clean
+	return nil
+}
+
+type scoreResult struct {
+	Title string  `json:"title"`
+	Score float64 `json:"score"`
+}
+
+func (c *ScoreCommand) Run(stdin io.Reader, stdout io.Writer) error {
+	model, err := loadModel(c.ModelPath)
+	if err != nil {
+		return err
+	}
+	vectorizer := core.CreateVectorizerFromModel(model)
+
+	enc := json.NewEncoder(stdout)
+	scanner := bufio.NewScanner(stdin)
+	scanner.Buffer(make([]byte, 64*1024), 1024*1024)
+	for scanner.Scan() {
+		title := strings.TrimSpace(scanner.Text())
+		if title == "" {
+			continue
+		}
+		if len(title) > 1000 {
+			return fmt.Errorf("title exceeds 1000 chars")
+		}
+		vec := vectorizer.Transform([]string{title})
+		score, err := core.PredictScore(vec[0], model.Weights)
+		if err != nil {
+			return fmt.Errorf("scoring %q: %w", title, err)
+		}
+		if err := enc.Encode(scoreResult{Title: title, Score: score}); err != nil {
+			return err
+		}
+	}
+	return scanner.Err()
+}
+
+func loadModel(path string) (*core.ModelEnvelope, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, fmt.Errorf("opening model: %w", err)
+	}
+	defer func() { _ = f.Close() }()
+
+	var model core.ModelEnvelope
+	if err := json.NewDecoder(f).Decode(&model); err != nil {
+		return nil, fmt.Errorf("parsing model %s: %w", path, err)
+	}
+	if len(model.Weights) == 0 || len(model.OrderedVocab) == 0 {
+		return nil, fmt.Errorf("model %s is missing weights or vocabulary", path)
+	}
+	return &model, nil
+}