mirror of
https://github.com/triqbit/mt5-ai-xauusd-trader.git
synced 2026-11-02 06:56:15 +00:00
27ba944d1c
* feat: implement institutional RL evaluation framework This PR implements a comprehensive reinforcement learning evaluation framework designed for institutional-grade trading research. Key features: - **Advanced Stability Metrics**: Beyond Sharpe/Sortino, it includes VaR, CVaR, Ulcer Index, SQN, Lake Ratio, and Tail Ratio. - **Trade Efficiency**: Calculates Maximum Adverse Excursion (MAE) and Maximum Favorable Excursion (MFE) for every trade. - **Turnover & Consistency**: Tracks trade frequency, hold times, action entropy, and a 'flip-flop' rate to detect excessive reversals. - **Market Awareness**: Segmented performance by institutional market regimes and trading sessions (Asian, London, NY). - **Statistical Rigor**: Integrates p-value calculations (paired T-test) to compare RL agents against rule-based baselines. - **Robust Trade Extraction**: Correctly handles position reversals and positions open at the start of evaluation. - **Typed Reporting**: Fully integrated with the project's reporting framework via Pydantic models and Jinja2 templates. Verified by 32 unit tests. Co-authored-by: saysgrok <279469430+saysgrok@users.noreply.github.com> * feat: implement institutional RL evaluation framework This PR implements a comprehensive reinforcement learning evaluation framework designed for institutional-grade trading research. Key features: - **Advanced Stability Metrics**: Beyond Sharpe/Sortino, it includes VaR, CVaR, Ulcer Index, SQN, Lake Ratio, and Tail Ratio. - **Trade Efficiency**: Calculates Maximum Adverse Excursion (MAE) and Maximum Favorable Excursion (MFE) for every trade. - **Turnover & Consistency**: Tracks trade frequency, hold times, action entropy, and a 'flip-flop' rate to detect excessive reversals. - **Market Awareness**: Segmented performance by institutional market regimes and trading sessions (Asian, London, NY). - **Statistical Rigor**: Integrates p-value calculations (paired T-test) to compare RL agents against rule-based baselines. - **Robust Trade Extraction**: Correctly handles position reversals and positions open at the start of evaluation. - **Typed Reporting**: Fully integrated with the project's reporting framework via Pydantic models and Jinja2 templates. Also includes dependency harmonization for python-socketio and various linting fixes to ensure CI passing. Co-authored-by: saysgrok <279469430+saysgrok@users.noreply.github.com> * feat: institutional rl evaluation framework Implemented comprehensive reinforcement learning evaluation logic in `src/research/rl_evaluation.py` with disciplined metrics beyond reward. Key features: - Stability analysis: Added MAE (Maximum Adverse Excursion) and MFE (Maximum Favorable Excursion) tracking. - Turnover analysis: Implemented a `flip_flop_rate` to detect rapid Buy/Sell oscillations. - Session attribution: Added PnL breakdown for Asian, London, and New York sessions (UTC) with session diversification entropy. - Statistical rigor: Integrated paired t-tests to calculate p-values relative to baselines. - Enhanced trade extraction: Robust processing for positions open at step 0 and direct Long-to-Short reversals. - Institutional reporting: Updated `src/research/reporting.py` models and templates to support the new metrics. - Environment harmonization: Aligned `python-socketio` to 5.14.0 for security and added `scipy` dependency. Verified with 32 unit tests in `tests/test_rl_evaluation.py` and 19 reporting tests. Co-authored-by: saysgrok <279469430+saysgrok@users.noreply.github.com> --------- Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com> Co-authored-by: saysgrok <279469430+saysgrok@users.noreply.github.com>
8 lines
151 B
Python
8 lines
151 B
Python
try:
|
|
import torch
|
|
except ImportError:
|
|
torch = None
|
|
import pytest
|
|
|
|
pytestmark = pytest.mark.skipif(torch is None, reason="torch not installed")
|