FDN-01: repository & projektgerüst

Git-Repository für bestehenden archivdms-Code initialisiert, Branch-/Commit-Konvention (feature/<ticket>-<slug>-Branches, Ticket-Prefix in Commit-Nachricht) etabliert.

internal/storage/ml_classifier_train.go

@@ -0,0 +1,59 @@
+package storage
+
+import (
+	"context"
+	"fmt"
+
+	"archivdms/internal/classifier"
+)
+
+// MLClassifierKinds is the fixed set of taxonomy kinds the Naive-Bayes
+// classifier is trained/predicted for, in a stable order (used by the retrain
+// CLI so runs are deterministic).
+var MLClassifierKinds = []string{
+	classifier.KindDocumentTypes,
+	classifier.KindCorrespondents,
+	classifier.KindTags,
+}
+
+// TrainClassifier rebuilds the Naive-Bayes model for one tenant and one kind and
+// returns the number of training documents used. Thin wrapper around
+// classifier.Train that keeps the Store's *pgxpool.Pool encapsulated (the
+// classifier package must not import storage). See classifier.Train for
+// semantics (full rebuild, MinDocsPerClass threshold, no error on empty data).
+func (s *Store) TrainClassifier(ctx context.Context, tenantID int64, kind string) (int, error) {
+	return classifier.New(s.db).Train(ctx, tenantID, kind)
+}
+
+// StartMLRun inserts a fresh ml_classifier_runs row in status 'running' for a
+// tenant and returns its id. The retrain CLI creates one run per tenant that
+// spans all kinds, then finalises it with FinishMLRun.
+func (s *Store) StartMLRun(ctx context.Context, tenantID int64) (int64, error) {
+	var id int64
+	if err := s.db.QueryRow(ctx,
+		`INSERT INTO ml_classifier_runs (tenant_id, status) VALUES ($1, 'running') RETURNING id`,
+		tenantID).Scan(&id); err != nil {
+		return 0, fmt.Errorf("storage: start ml_classifier run: %w", err)
+	}
+	return id, nil
+}
+
+// FinishMLRun finalises an ml_classifier_runs row: sets completed_at=now(),
+// doc_count, the terminal status ('completed', 'failed' or
+// 'skipped_insufficient_data') and an optional error message. errMsg "" stores
+// SQL NULL.
+func (s *Store) FinishMLRun(ctx context.Context, runID int64, docCount int, status, errMsg string) error {
+	var errArg any
+	if errMsg != "" {
+		errArg = errMsg
+	}
+	_, err := s.db.Exec(ctx,
+		`UPDATE ml_classifier_runs
+		 SET completed_at = now(), doc_count = $2, status = $3, error = $4
+		 WHERE id = $1`,
+		runID, docCount, status, errArg)
+	if err != nil {
+		return fmt.Errorf("storage: finish ml_classifier run: %w", err)
+	}
+	return nil
+}