sketch_naive_bayes.nx

sketch_naive_bayes.nx -- streaming binary Naive Bayes classifier. Online supervised-learning primitive. Maintains per-class feature counts and class priors. Classify in log-space via: score(class) = log P(class) + Σ_f log P(f | class) predict = arg max class { score(class) } where: P(class) = class_count / total_obs P(feature|class) = (feature_class_count + alpha) / (class_count + alpha * V) alpha = 1 for Laplace smoothing (avoids -inf for unseen features) V = vocabulary size (caller's responsibility to bound) LOG-SPACE PROBABILITIES (integer): log_2(p) approximated via bitlen(p) - 1 + linear fractional bits. Sum-of-logs in PPM. COMPOSES against sketch_hash_map for sparse feature storage. USE CASES: - spam vs ham email classification - log-line severity (error/warning/info) - sentiment (positive/negative) - URL category (safe / suspicious) API: train(class, feature_id) // observe (class, feature) pair predict(features, n_features) -> class label (0 or 1) score(class, features, n) -> log-prob in PPM LOSSLESS-LANGUAGE DISCIPLINE: Prediction is a sealed binary value (0 or 1). Posterior probability returned in PPM with NX_ENV_REL_STDDEV.