Models Overview

All 7 models implement the BaseModel abstract interface and are accessible through the MODEL_REGISTRY.

Model Comparison

Model	Type	Key Config	Strengths
Logistic Regression	Tabular	C=1.0, balanced weights	Fast, interpretable baseline
Random Forest	Tabular	200 trees, max_depth=15	Handles non-linear relationships
HMM	Sequence	1-8 hidden states	Captures latent pitch states
AutoGluon	Tabular	good_quality preset	Automated model selection
LSTM	Sequence	2-layer, hidden=64, window=8	Long-range sequence dependencies
1D-CNN	Sequence	3 conv layers, kernel=3	Local pattern detection
Transformer	Sequence	d_model=64, 4 heads, 2 layers	Self-attention over sequences

BaseModel Interface

All models implement the following abstract interface:

from pitch_sequencing.models.base import BaseModel

class BaseModel(ABC):
    @property
    def name(self) -> str:
        """Human-readable model name."""

    @property
    def model_type(self) -> str:
        """'tabular' or 'sequence' — determines input shape."""

    def fit(self, X_train, y_train, X_val=None, y_val=None, **kwargs):
        """Train the model."""

    def predict(self, X) -> np.ndarray:
        """Return predicted class labels."""

    def predict_proba(self, X) -> np.ndarray:
        """Return class probabilities (n_samples x n_classes)."""

    def get_params(self) -> dict:
        """Return model hyperparameters."""

Input Shapes

• Tabular models (model_type = "tabular"): Expect (n_samples, n_features) NumPy arrays or DataFrames
• Sequence models (model_type = "sequence"): Expect (n_samples, window_size, n_features) 3D arrays

Model Registry

Models are accessed by name through the registry:

from pitch_sequencing import get_model, MODEL_REGISTRY

# List all registered models
print(list(MODEL_REGISTRY.keys()))
# ['logistic_regression', 'random_forest', 'hmm', 'autogluon', 'lstm', 'cnn1d', 'transformer']

# Instantiate a model with config
model = get_model("lstm", {"hidden_size": 64, "num_layers": 2})