superneuromat.SNN

class superneuromat.SNN

Parameters:

• num_neurons (int) – The number of neurons in the SNN.

• neurons (NeuronList) – List of neurons in the SNN.

• num_synapses (int) – The number of synapses in the SNN.

• synapses (SynapseList) – List of synapses in the SNN.

• spike_train (list) – List of output spike trains for each time step. Index by spike_train[t][neuron_id].

• ispikes (numpy.ndarray[(int, int), bool]) – Binary numpy array of input spikes for each time step. Index by ispikes[t, neuron_id].

STDP Parameters

Parameters:

• stdp (bool, default=True) – if stdp is False, STDP will be disabled globally.

• apos (list, default=[]) – List of STDP parameters per time step for excitatory update of weights.

• aneg (list, default=[]) – List of STDP parameters per time step for inhibitory update of weights.

• stdp_positive_update (bool | Any, default=True) – If False, disable excitatory STDP update globally.

• stdp_negative_update (bool | Any, default=True) – If False, disable inhibitory STDP update globally.

• stdp_time_steps (int) – Number of time steps over which STDP updates will be made. This is determined from the length of apos or aneg.

Config Parameters

Parameters:

• default_dtype (type | numpy.dtype, default=numpy.float64) – Default data type for internal numpy representation of all floating-point arrays.

• default_bool_dtype (type | numpy.dtype, default=bool) – Default data type for internal numpy representation of stdp_enabled values and spike vectors.

• gpu (bool) – If False, disable GPU acceleration. By default, this is True if numba.cuda.is_available().

• manual_setup (bool, default=False) – If True, disable automatic setup of SNN when simulate() is called.

• allow_incorrect_stdp_sign (bool, default=False) – If True, allow negative values in apos and positive values in aneg.

• allow_signed_leak (bool, default=False) – If True, allow negative values in neuron_leaks. Normally, positive leak will result in charge decay.

• backend (str, default='auto') – The backend to use for simulation. Can be 'auto', 'cpu', 'jit', or 'gpu'. Can be overridden on a per-simulate() basis.

• sparse (bool | str | Any, default=False) – If True, use sparse representation of SNN. If 'auto', use sparse representation if num_neurons > 100. Sparse representation is only available using the 'cpu' backend.

• last_used_backend (str, default=None) – The backend most recently used for simulation.

• is_sparse (bool, default=False) – If True, the SNN is being internally represented using a sparse representation.

Data Attributes

These attributes are not intended to be modified directly by end users.

Parameters:

• neuron_thresholds (list) – List of neuron thresholds

• neuron_leaks (list) – List of neuron leaks defined as the amount by which the internal states of the neurons are pushed towards the neurons’ reset states

• neuron_reset_states (list) – List of neuron reset states

• neuron_refractory_periods (list) – List of neuron refractory periods

• pre_synaptic_neuron_ids (list) – List of pre-synaptic neuron IDs

• post_synaptic_neuron_ids (list) – List of post-synaptic neuron IDs

• synaptic_weights (list) – List of synaptic weights

• synaptic_delays (list) – List of synaptic delays

• enable_stdp (list) – List of Boolean values denoting whether STDP learning is enabled on each synapse

• input_spikes (dict) – Dictionary of input spikes for each time step.

Warning

Delay is implemented by adding a chain of proxy neurons.

A delay of 10 between neuron A and neuron B would add 9 proxy neurons between A and B. This may result in severe performance degradation. Consider using sparse representation or agent-based SNN simulators in networks with high delay times.

Hint

• Leak brings the internal state of the neuron closer to the reset state.

  The leak value is the amount by which the internal state of the neuron is pushed towards its reset state.

• Deletion of neurons may result in unexpected behavior.

• Input spikes can have a value

• All neurons are monitored by default

__init__()

Initialize the SNN

Methods

SNN.__init__()

Initialize the SNN

Building the SNN

add_spike	Adds an external spike in the SNN
create_neuron	Create a neuron in the SNN.
create_synapse	Creates a synapse in the SNN
stdp_setup	Setup the Spike-Time-Dependent Plasticity (STDP) parameters
set_stdp_enabled_from_mat
set_weights_from_mat	Set the synaptic weights from a matrix.

Inspecting the SNN

get_neuron_df	Returns a DataFrame containing information about neurons.
neuron_spike_totals	Returns a vector of the number of spikes emitted from each neuron since last reset.
get_synapse	Returns the synapse that connects the given pre- and post-synaptic neurons.
get_synapse_df	Returns a DataFrame containing information about synapses.
get_synapses_by_post	Returns the synapses that connect the given post-synaptic neuron.
get_synapses_by_pre	Returns a list of synapses with the given pre-synaptic neuron.
get_synapse_id	Returns the id of the synapse connecting the given pre- and post-synaptic neurons.
get_synaptic_ids_by_post	Returns a list of synapse ids with the given post-synaptic neuron.
get_synaptic_ids_by_pre	Returns a list of synapse ids with the given pre-synaptic neuron.
get_weights_df	Returns a DataFrame containing information about synapse weights.
input_spikes_info	Generate a description of the synapses in the SNN.
get_input_spikes_df	Returns a DataFrame containing information about input spikes.
get_stdp_enabled_df	Returns a DataFrame containing information about STDP enabled synapses.
last_used_backend	Get the last-used backend for simulation.
pretty	Generates a text description of the model.
pretty_print	Pretty-prints the model.
pretty_spike_train	Returns a pretty string of the spike train.
print_spike_train	Prints the spike train.
short	Return a 1-line summary of the SNN.
neuron_info	Generate a description of the neurons in the SNN.
stdp_info	Generate a description of the current global STDP settings.
synapse_info	Generate a description of the synapses in the SNN.
stdp_enabled_mat	Create a boolean dense matrix which indicates whether STDP is enabled on each synapse.
stdp_enabled_sparse
weight_mat	Create a dense weight matrix from the synaptic weights.
weight_sparsity
weights_sparse	Create a sparse weight matrix from the synaptic weights.

Managing Model State

clear_input_spikes
clear_memos	Delete all stored copies of model variables.
clear_spike_train
consume_input_spikes	Consumes/deletes input spikes for the given number of time steps.
copy	Returns a copy of the SNN
devec	Copy the internal state variables back to the public-facing canonical representations.
memoize	Store a copy of model variable(s) to be restored later.
recommend	Recommend a backend to use based on network size and continuous sim time steps.
recommend_sparsity
release_mem	Delete internal variables created during computation.
reset	Reset the SNN's neuron states, refractory periods, spike train, and input spikes.
reset_neuron_states
reset_refractory_periods
restore	Restore model variables to their memoized states.
setup	Setup the SNN for simulation.
setup_input_spikes
simulate	Simulate the neuromorphic spiking neural network
simulate_cpu
simulate_cpu_jit
simulate_gpu	Simulate the neuromorphic circuit using the GPU backend.
unmemoize	Delete stored copies of model variables.
zero_neuron_states
zero_refractory_periods
rebuild_connection_ids
to_json	Exports the SNN to a JSON string.
saveas_json	Exports the SNN to a JSON file.
from_jsons	Update this SNN from a SuperNeuroMat JSON string.

Attributes

Attributes

num_neurons	Returns the number of neurons in the SNN.
num_synapses	Returns the number of synapses in the SNN.
ispikes	Convert the output spike train to a dense binary numpy.ndarray.
backend	Set the backend to be used for simulation.
disable_performance_warnings
eqvars	The list of public variables which define the SNN model.
_internal_vars	NumPy representation of the internal state variables used during simulation.
is_sparse	Returns True if SNN is sparse internally.
sparse	The user-requested sparsity setting for the SNN
stdp_time_steps	Returns the number of time steps over which STDP updates will be made.
jit_threshold
gpu_threshold
sparsity_threshold