v0.2.2

What's New

• MLP Hyperparameter Optimization: --optimize flag in train_reranker.py uses RandomizedSearchCV to search over architecture, learning rate, alpha, activation, and max_iter
• Per-Domain Threshold Tuning: --optimize-thresholds flag finds optimal mlp_uncertainty_low/high values per domain, maximizing accuracy + coverage
• New training params: alpha, learning_rate_init, activation exposed in both MLPReranker.train() and arf.trainer.train_reranker() public API
• Config schema validation: New MLP fields (mlp_alpha, mlp_learning_rate_init, mlp_hidden_layer_sizes, mlp_activation, mlp_max_iter) in config_schema.py

Install

pip install advanced-rag-framework==0.2.2