spires_compute_output

Compute the readout output vector from the reservoir’s current state: y = W_out * x.

Signature

spires_status spires_compute_output(spires_reservoir *r, double *out);

Parameters

Returns

spires_status — SPIRES_OK on success, or SPIRES_ERR_INVALID_ARG if r or out is NULL.

Example

size_t no = spires_num_outputs(r);
double *y = malloc(no * sizeof(double));
 
/* Step and read output */
spires_step(r, input_vec);
spires_compute_output(r, y);
 
for (size_t k = 0; k < no; k++) {
    printf("output[%zu] = %.6f\n", k, y[k]);
}
 
free(y);

Notes

• Readout weights. The output is the matrix-vector product W_out * x, where W_out is the [num_outputs x num_neurons] readout weight matrix and x is the current neuron state vector. If no training has been performed, W_out is all zeros and the output will be zero.
• No allocation. This function performs no heap allocation. The caller provides the output buffer.
• Buffer sizing. The out array must hold at least num_outputs doubles. Use spires_num_outputs to query the required size.
• Idempotent. Calling this function does not modify the reservoir state. It can be called multiple times between steps with identical results.
• Thread safety. Safe to call concurrently with other read-only functions (e.g., spires_num_neurons) on the same reservoir, but do not call concurrently with mutating functions (e.g., spires_step, spires_train_online).

See Also

• spires_step — advance the state before computing output.
• spires_train_ridge — train the output weights.
• spires_train_online — update the output weights incrementally.
• spires_num_outputs — query the output dimension.