What problem does it solve? Validating whether a parameter change (holding period, score threshold, risk control) actually improves a quantitative trading strategy requires a full train-infer-backtest-report loop; this Skill automates that entire closed loop on the QuantMind platform so decisions are driven by data instead of guesswork. ## Core Features & Use Cases - Model Training: Submit T+N cycle training jobs for 13 model types (lightgbm, xgboost, catboost, random_forest, linear, mlp, gru, lstm, alstm, transformer, tabnet, tcn, nativetft) with GPU training and quality gates (Rank IC/ICIR). - Batch Inference: Run range-mode batch inference across a full year of trading days, writing daily scores to the signal database. - Portfolio Backtesting: Execute optimized backtests with score thresholds, market-index MA filtering, 5% stop-loss, slippage, T+1, and ST-stock exclusion. - Report Generation: Produce research-grade Markdown reports and convert them to styled PDFs. - Use Case: Compare a T+3 model against an existing T+5 model by cloning the training payload, changing only the horizon, running full-year inference and identical backtests, then exporting a comparison report. ## Quick Start Train a T+3 CatBoost model cloned from my existing T+5 job, run full-year batch inference, backtest it with the optimized strategy, and generate a PDF report comparing the two periods.
