ci: add dedicated PPO hyperopt training job
Adds hyperopt-ppo job (manual, 4h timeout) that runs: 1. PSO hyperopt (20 trials, 15 epochs each) to find optimal PPO params 2. Full training with best params 3. Walk-forward evaluation with Sharpe/MaxDD/WinRate metrics Artifacts: ppo_hyperopt_results.json + ppo_eval_report.json (90 days) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
@@ -510,6 +510,44 @@ train-validate-tft:
|
||||
when: always
|
||||
expire_in: 30 days
|
||||
|
||||
hyperopt-ppo:
|
||||
extends: .train-validate-base
|
||||
needs: [build-training]
|
||||
timeout: 4h
|
||||
script:
|
||||
- /usr/local/bin/hyperopt_baseline_rl
|
||||
--model ppo
|
||||
--trials 20
|
||||
--n-initial 5
|
||||
--epochs 15
|
||||
--symbol ES.FUT
|
||||
--data-dir /mnt/training-data/futures-baseline
|
||||
--base-dir ${CI_PROJECT_DIR}/output/hyperopt
|
||||
--output ${CI_PROJECT_DIR}/output/ppo_hyperopt_results.json
|
||||
- echo "=== PPO Hyperopt complete ==="
|
||||
- cat ${CI_PROJECT_DIR}/output/ppo_hyperopt_results.json
|
||||
- echo "=== Training PPO with best params ==="
|
||||
- /usr/local/bin/train_baseline_rl
|
||||
--model ppo
|
||||
--symbol ES.FUT
|
||||
--data-dir /mnt/training-data/futures-baseline
|
||||
--output-dir ${CI_PROJECT_DIR}/output
|
||||
--hyperopt-params ${CI_PROJECT_DIR}/output/ppo_hyperopt_results.json
|
||||
- echo "=== Evaluating PPO with best params ==="
|
||||
- /usr/local/bin/evaluate_baseline
|
||||
--model ppo
|
||||
--models-dir ${CI_PROJECT_DIR}/output
|
||||
--data-dir /mnt/training-data/futures-baseline
|
||||
--output ${CI_PROJECT_DIR}/output/ppo_eval_report.json
|
||||
--symbol ES.FUT
|
||||
- cat ${CI_PROJECT_DIR}/output/ppo_eval_report.json
|
||||
artifacts:
|
||||
paths:
|
||||
- output/ppo_hyperopt_results.json
|
||||
- output/ppo_eval_report.json
|
||||
when: always
|
||||
expire_in: 90 days
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# IaC: Terragrunt plan on MR (runs on gitlab pool)
|
||||
# --------------------------------------------------------------------------
|
||||
|
||||
Reference in New Issue
Block a user