Function _extract_all_features

uncommon_route/router/classifier.py:129–165

Extract the complete feature vector — no keywords. Structural features detect code/math/structure via symbols and character-level statistics. N-grams learn vocabulary-equivalent patterns from training data. Context features encode agentic step information as pure numerical signals

(
    prompt: str,
    system_prompt: str | None = None,
    context_features: dict[str, float] | None = None,
)

def _extract_all_features(
    prompt: str,
    system_prompt: str | None = None,
    context_features: dict[str, float] | None = None,
) -> dict[str, float]:
    """Extract the complete feature vector — no keywords.

    Structural features detect code/math/structure via symbols and
    character-level statistics.  N-grams learn vocabulary-equivalent
    patterns from training data.  Context features encode agentic
    step information as pure numerical signals.
    """
    feature_text = _merge_feature_text(prompt, system_prompt)
    struct_dims = extract_structural_features(feature_text)
    structural_scores = {d.name: d.score for d in struct_dims}

    unicode_blocks = extract_unicode_block_features(feature_text)

    if _model is not None:
        return _model._build_features(
            structural_scores,
            unicode_blocks,
            keyword_scores=None,
            prompt=feature_text,
            context_features=context_features,
        )

    features: dict[str, float] = {}
    for name, score in structural_scores.items():
        features[f"s_{name}"] = score
    for name, prop in unicode_blocks.items():
        features[f"u_{name}"] = prop
    if context_features:
        for name, value in context_features.items():
            key = name if name.startswith("ctx_") else f"ctx_{name}"
            features[key] = value
    return features


def train_and_save_model(data_path: str, out_path: str | None = None) -> None: