Module decoding 

Functions

bayesian_decode

Bayesian MAP decoder (Dayan & Abbott 2001). spike_counts: (n_neurons,). tuning_rates: (n_stimuli × n_neurons, row-major flat). prior: (n_stimuli,) or empty for uniform. Returns MAP stimulus index.

linear_discriminant_decode

Fisher linear discriminant decoder (Fisher 1936). train_data: (n_samples × n_features, row-major flat). labels: (n_samples,). test_point: (n_features,). Returns predicted class label.

maximum_likelihood_decode

Maximum likelihood stimulus decoder (Dayan & Abbott 2001). Uniform prior.

naive_bayes_decode

Gaussian naive Bayes decoder (Mitchell 1997). train_data: (n_samples × n_features, row-major flat). labels: (n_samples,). test_point: (n_features,). Returns predicted class label.

population_vector_decode

Georgopoulos population vector decoding. trains: slice of binary spike trains (i32). preferred_directions: angle per neuron (radians). Returns decoded angle per time bin.

solve_spd 🔒

Solve the symmetric positive-definite system A X = B via Cholesky factorisation. a is n×n row-major — here the ridge-regularised within-class scatter, which is SPD — and b is n×m row-major; returns X (n×m, row-major). A single factorisation solves every right-hand-side column (one per class), the numerically optimal route for an SPD system: half the arithmetic of a general elimination and unconditionally stable without pivoting. Falls back to a zero solution if a is not positive-definite, which the ridge precludes.