spires_step

Advance the reservoir by a single timestep, feeding in one input vector.

Signature

spires_status spires_step(spires_reservoir *r, const double *u_t);

Parameters

Returns

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

Example

/* Step through a time series one sample at a time */
for (size_t t = 0; t < series_length; t++) {
    const double *u = &input_series[t * num_inputs];
    spires_status s = spires_step(r, u);
    if (s != SPIRES_OK) {
        fprintf(stderr, "step failed at t=%zu\n", t);
        break;
    }
 
    /* Read out the current state or output after each step */
    double y;
    spires_compute_output(r, &y);
    printf("t=%zu  y=%.6f\n", t, y);
}

Notes

• State update. Each call updates the internal neuron state vector according to the reservoir’s neuron model: x(t+1) = f(W * x(t) + W_in * u(t)). The exact dynamics depend on the neuron_type selected at creation time.
• NULL input. Passing u_t = NULL is equivalent to passing a zero vector. This is useful for “free-running” the reservoir without external drive, for example to observe the decay of transient dynamics.
• No allocation. This function performs no heap allocation. It operates entirely on pre-allocated internal buffers.
• Thread safety. A given reservoir must not be stepped from multiple threads simultaneously. Different reservoirs may be stepped concurrently without issue.
• Fractional models. For FLIF_* neuron types, each call to spires_step appends to the internal history buffer used for the fractional derivative computation. The memory cost per step is O(1) for FLIF_GL and FLIF_DIFFUSIVE, but may vary for FLIF_CAPUTO.

See Also

• spires_run — drive the reservoir with an entire series in one call.
• spires_compute_output — read the output after stepping.
• spires_copy_reservoir_state — snapshot the state after stepping.