Найдите подходящую возможность для вашего агента.
Model-trader version compatibility protocol: Embed version metadata in checkpoints, validate at load time. Trigger when: (1) training and live trading versions diverge, (2) models fail to load, (3) action interpretation issues.
Systematically search hyperparameter space. Use when tuning learning rate, batch size, or other hyperparameters.
Debug TensorFlow and Keras issues systematically. This skill helps diagnose and resolve machine learning problems including tensor shape mismatches, GPU/CUDA detection failures, out-of-memory errors, NaN/Inf values in loss functions, vanishing/exploding gradients, SavedModel loading errors, and data pipeline bottlenecks. Provides tf.debugging assertions, TensorBoard profiling, eager execution debugging, and version compatibility guidance.
7-action space with position sizing (25/50/75%) + small account simulation. Trigger when: (1) model needs sizing decisions, (2) training for <$25K accounts, (3) upgrading obs_dim 5600->5900.
Rigorous A/B/C testing framework for empirically evaluating reasoning patterns. Use when you need data-driven pattern selection, want to quantify trade-offs between patterns, or need to validate claims about which cognitive methodology performs best. Enables scientific measurement of quality, cost, and time trade-offs across ToT, BoT, SRC, HE, AR, DR, AT, RTR, and NDF patterns.
Add account state (P&L, win rate, drawdown) to RL observations + drawdown penalty in rewards. Trigger when: (1) model needs account awareness, (2) training should penalize drawdowns, (3) upgrading obs_dim 5300→5600.
Run Microsoft's eval-recipes benchmarks to validate amplihack improvements against baseline agents. Activates when testing with eval-recipes, running evals, or benchmarking changes.
Decomposed reasoning with explicit confidence scoring. Use for complex decisions, debugging failures, and architectural choices where tracking uncertainty prevents wasted effort.
Train RL models across multiple timeframes with resampling. Trigger when: (1) multi-timeframe training, (2) resampling data, (3) creating 1Hour/4Hour models.
Debug Scikit-learn issues systematically. Use when encountering model errors like NotFittedError, shape mismatches between train and test data, NaN/infinity value errors, pipeline configuration issues, convergence warnings from optimizers, cross-validation failures due to class imbalance, data leakage causing suspiciously high scores, or preprocessing errors with ColumnTransformer and feature alignment.
Detects unsafe training load spikes (>20-30% week-over-week) and emits safety flags. Use in nightly background jobs or when reviewing weekly training volume with conservative adjustment recommendations.
Run and monitor ATFT-GAT-FAN training loops, hyper-parameter sweeps, and safety modes on A100 GPUs.
Interactive diagnostic workflow for training problems. Use when training is failing, loss is stuck, gradients explode, NaN occurs, or convergence is poor.
Implement gradient accumulation that produces bit-identical results to standard batching. Use when: comparing GA vs non-GA runs, debugging training reproducibility.
Python best practices with modern frameworks, AI/ML integration, and performance optimization for 2025
