`spires_reservoir_config`

Struct that fully describes a reservoir before it is created. Passed by pointer to spires_reservoir_create.

Definition


typedef struct {
    size_t num_neurons;
    size_t num_inputs;
    size_t num_outputs;
    double spectral_radius;
    double ei_ratio;
    double input_strength;
    double connectivity;
    double dt;
    spires_connectivity_type connectivity_type;
    spires_neuron_type       neuron_type;
    double *neuron_params;
} spires_reservoir_config;

Fields

Field	Type	Description	Typical Values
`num_neurons`	`size_t`	Number of neurons in the reservoir. Determines the dimensionality of the internal state vector.	100 — 5000
`num_inputs`	`size_t`	Number of input channels. Each call to `spires_step` expects a vector of this length.	1 — 128
`num_outputs`	`size_t`	Number of output channels. Determines the width of the readout weight matrix `W_out`.	1 — 64
`spectral_radius`	`double`	Desired spectral radius of the internal weight matrix. Controls the echo state property: values below 1.0 ensure fading memory, values near 1.0 maximize memory capacity.	0.1 — 1.5 (commonly 0.9)
`ei_ratio`	`double`	Fraction of neurons designated as excitatory, in the range [0, 1]. The remainder are inhibitory. A value of 0.8 means 80% excitatory, 20% inhibitory.	0.5 — 1.0 (commonly 0.8)
`input_strength`	`double`	Scaling factor applied to the input weight matrix `W_in`. Larger values amplify the influence of external input relative to recurrent dynamics.	0.01 — 5.0 (commonly 1.0)
`connectivity`	`double`	Controls how densely the reservoir is connected. For `SPIRES_CONN_RANDOM`, this is the probability that any two neurons are connected (range [0, 1]). For other topologies, its interpretation varies (see Enums).	0.01 — 0.5 (commonly 0.1)
`dt`	`double`	Integration timestep in seconds. Used by biophysical and fractional neuron models. Ignored by `SPIRES_NEURON_LIF_DISCRETE`.	0.0001 — 0.01
`connectivity_type`	`spires_connectivity_type`	Network topology algorithm. See `spires_connectivity_type`.	`SPIRES_CONN_RANDOM`
`neuron_type`	`spires_neuron_type`	Spiking neuron model. See `spires_neuron_type`.	`SPIRES_NEURON_LIF_DISCRETE`
`neuron_params`	`double *`	Pointer to a model-specific parameter array. May be `NULL` for models that require no extra parameters (e.g., `LIF_DISCRETE`). For fractional models, the first element is the fractional order alpha. The library reads from this pointer during creation but does not take ownership.	See notes below

`neuron_params` Layout

The contents of the neuron_params array depend on neuron_type:

Neuron Type	Required Elements	Layout
`SPIRES_NEURON_LIF_DISCRETE`	0	`NULL` or ignored.
`SPIRES_NEURON_LIF_BIO`	4	`[tau_m, V_rest, V_thresh, V_reset]` — membrane time constant (s), resting potential (V), threshold (V), reset potential (V).
`SPIRES_NEURON_FLIF_CAPUTO`	1+	`[alpha, ...]` — fractional order in (0, 1]. Additional elements are model-specific.
`SPIRES_NEURON_FLIF_GL`	1+	`[alpha, ...]` — fractional order in (0, 1].
`SPIRES_NEURON_FLIF_DIFFUSIVE`	1+	`[alpha, ...]` — fractional order in (0, 1].

Example


double frac_params[] = { 0.7 };  /* fractional order alpha = 0.7 */
 
spires_reservoir_config cfg = {
    .num_neurons      = 500,
    .num_inputs       = 3,
    .num_outputs      = 1,
    .spectral_radius  = 0.95,
    .ei_ratio         = 0.8,
    .input_strength   = 1.0,
    .connectivity     = 0.1,
    .dt               = 0.001,
    .connectivity_type = SPIRES_CONN_SMALL_WORLD,
    .neuron_type       = SPIRES_NEURON_FLIF_GL,
    .neuron_params     = frac_params
};

Notes

The config struct is read-only from the library’s perspective. spires_reservoir_create copies all necessary data out of the struct and the neuron_params array. After creation, the caller may freely modify or deallocate the config and its neuron_params.
Zero-initializing the struct (e.g., memset(&cfg, 0, sizeof(cfg))) is not sufficient for a valid configuration because num_neurons, num_inputs, and num_outputs must all be non-zero.
The spectral_radius field may be set to 0 to skip spectral rescaling of the internal weight matrix. This is useful when providing a custom pre-scaled matrix through advanced APIs.

spires_reservoir_config

Definition

Fields

neuron_params Layout

Example

Notes

See Also

`spires_reservoir_config`

`neuron_params` Layout