Skip to main content

ML Train Latest Select

npx @tradejs/cli ml-train:latest runs interactive/selective training from latest exports and auto-prepares holdout/prod/walk-forward splits.

Runtime surface:

  • @tradejs/cli
  • your configured ML training runtime (container/service)

What It Does​

  1. Select strategy and model type.
  2. Pick latest dataset (or merge multiple base exports when configured).
  3. Build derived split files:
  • *.holdout-train.<key>.jsonl
  • *.holdout-test.<key>.jsonl
  • *.prod.<key>.jsonl
  • *.walk-forward-fold-<N>.train.<key>.jsonl
  • *.walk-forward-fold-<N>.test.<key>.jsonl
  1. Cache split results by deterministic key/hash.
  2. Enforce causality guard on timestamp-like features.
  3. Start the configured ML training runtime.

Commands​

npx @tradejs/cli ml-train:latest
npx @tradejs/cli ml-train:latest --strategy TrendLine --model xgboost

Non-interactive options:

  • --strategy
  • --model (catboost|random_forest|extra_trees|xgboost|lightgbm)
  • --latestOnly

Main Environment Controls​

Windows and splits:

  • ML_TRAIN_RECENT_DAYS
  • ML_TRAIN_TEST_DAYS
  • ML_TRAIN_WALK_FORWARD_FOLDS

Ensemble/profile:

  • ML_TRAIN_ENSEMBLE
  • ML_TRAIN_FORCE_ENSEMBLE
  • ML_TRAIN_NO_ENSEMBLE
  • ML_TRAIN_FEATURE_PROFILE

Causality:

  • ML_TRAIN_DISABLE_CAUSALITY_GUARD=1 (debug only)

Incremental mode:

  • ML_TRAIN_INCREMENTAL_THRESHOLD_GB
  • ML_TRAIN_INCREMENTAL
  • ML_TRAIN_NO_INCREMENTAL
  • ML_TRAIN_CHUNK_SIZE
  • ML_TRAIN_INCREMENTAL_ITERATIONS

Logging/debug:

  • ML_TRAIN_DEBUG=1
  • ML_TRAIN_HEARTBEAT_SEC
  • ML_TRAIN_DOCKER_NO_OUTPUT_TIMEOUT_SEC

Redis Safety Behavior​

In the reference deployment, training runs with a safety wrapper:

  • pause Redis before train
  • restore Redis on exit (success/failure/interrupt)

Output Artifacts​

  • model aliases (<Strategy>.joblib, ensemble aliases)
  • eval/prod snapshots
  • sidecar metrics JSON
  • markdown/html reports
  • split metadata (*.windows.<key>.meta.json)