spires_train_ridge

Train the readout weight matrix using batch ridge regression over a complete input/target time series.

Signature

spires_status spires_train_ridge(spires_reservoir *r,
                                 const double *input_series,
                                 const double *target_series,
                                 size_t series_length,
                                 double lambda);

Parameters

Returns

spires_status — SPIRES_OK on success. On failure:

• SPIRES_ERR_INVALID_ARG — a pointer is NULL, series_length is 0, or lambda is negative.
• SPIRES_ERR_ALLOC — memory allocation for internal work arrays failed.
• SPIRES_ERR_INTERNAL — the LAPACK linear solve failed (e.g., singular matrix).

Example

size_t T = 2000;
 
/* Drive reservoir to collect states */
/* (spires_train_ridge does this internally) */
 
spires_status s = spires_train_ridge(r, input, target, T, 1e-6);
if (s != SPIRES_OK) {
    fprintf(stderr, "training failed: %d\n", s);
    return 1;
}
 
/* Now compute outputs on new data */
spires_reservoir_reset(r);
for (size_t t = 0; t < test_len; t++) {
    spires_step(r, &test_input[t * num_inputs]);
    double y;
    spires_compute_output(r, &y);
    printf("%.6f\n", y);
}

Notes

• Internal workflow. This function (1) resets the reservoir state, (2) drives it with input_series to collect a state matrix X of shape [series_length x num_neurons], and (3) solves W_out = (X^T X + lambda * I)^{-1} X^T Y using LAPACK. The computed W_out is stored inside the reservoir for use by spires_compute_output.
• State after training. The reservoir state after this call reflects the final timestep of input_series. Call spires_reservoir_reset before processing new sequences.
• LAPACK dependency. This function requires a LAPACK implementation at link time (e.g., OpenBLAS, Apple Accelerate, Intel MKL). If LAPACK is not available, the function will return SPIRES_ERR_INTERNAL.
• Choosing lambda. Typical values range from 1e-8 to 1e-2. Cross-validation or the spires_optimize function can be used to select an appropriate value.
• Thread safety. Do not call this function on a reservoir that is concurrently in use by another thread.

See Also

• spires_train_online — incremental weight update.
• spires_compute_output — use the trained weights.
• spires_optimize — automated hyperparameter search including lambda.
• Ridge Regression — theoretical background.