ml_genn.callbacks package

Callbacks are used to run custom logic mid-simulation including for recording state.

class ml_genn.callbacks.BatchProgressBar

Bases: Callback

Callback to display a tqdm progress bar which gets updated every batch.

on_batch_end(batch, metrics)
on_epoch_begin(epoch)
on_test_begin()
on_test_end(metrics)
on_train_begin()
on_train_end(metrics)
set_params(num_batches, **kwargs)
class ml_genn.callbacks.Callback

Bases: object

Base class for all callbacks

class ml_genn.callbacks.Checkpoint(serialiser='numpy', epoch_interval=1)

Bases: Callback

Callback which serialises network state after a specified number of epochs.

Parameters:

  • serialiser (Serialiser | str) – Serialiser to use

  • epoch_interval (int) – After how many epochs should checkpoints be saved?

on_epoch_end(epoch, metrics)
set_params(compiled_network, **kwargs)
class ml_genn.callbacks.ConnVarRecorder(conn, genn_var, key=None, example_filter=None, src_neuron_filter=None, trg_neuron_filter=None)

Bases: Callback

Callback used for recording connection state variables during simulation. Variables are specified using GeNN state variable name. By convention, g is always the weight and d is the per-synapse delay if it is used. Other variables are compiler-specific e.g. Gradient accomulates gradients when using ml_genn.compilers.EventPropCompiler.

Parameters:

  • conn (Connection) – Synapse population to record from

  • genn_var (str) – Internal name of variable to record

  • key – Key to assign recording data produced by this callback in dictionary returned by evaluation/training methods of compiled network

  • example_filter (Sequence | ndarray | integer | int | None) – Filter used to select which examples to record from (see Callbacks and recording for more information).

  • src_neuron_filter (Sequence | slice | ndarray | integer | int | None) – Filter used to select which synapses to record from (see Callbacks and recording for more information).

  • trg_neuron_filter (Sequence | slice | ndarray | integer | int | None) – Filter used to select which synapses to record from (see Callbacks and recording for more information).

get_data()
on_batch_begin(batch)
Parameters:

batch (int)

on_timestep_end(timestep)
Parameters:

timestep (int)

set_params(data, compiled_network, **kwargs)
class ml_genn.callbacks.CustomUpdateOnBatchBegin(name)

Bases: CustomUpdate

Callback that triggers a GeNN custom update at the beginning of every batch.

Parameters:

name (str)

on_batch_begin(batch)
class ml_genn.callbacks.CustomUpdateOnBatchEnd(name)

Bases: CustomUpdate

Callback that triggers a GeNN custom update at the end of every batch.

Parameters:

name (str)

on_batch_end(batch, metrics)
class ml_genn.callbacks.CustomUpdateOnTimestepBegin(name)

Bases: CustomUpdate

Callback that triggers a GeNN custom update at the beginning of every timestep.

Parameters:

name (str)

on_timestep_begin(timestep)
class ml_genn.callbacks.CustomUpdateOnTimestepEnd(name)

Bases: CustomUpdate

Callback that triggers a GeNN custom update at the end of every timestep.

Parameters:

name (str)

on_timestep_end(timestep)
class ml_genn.callbacks.OptimiserParamSchedule(param_name, func)

Bases: Callback

Callback which updates an parameter on an optimisers.Optimiser every epoch based on a callable.

Parameters:

  • param_name (str) – Name of parameter to update. Not all optimiser parameters can be changed at runtime

  • func (Callable[[int, Number], Number]) – Callable called every epoch to determine new parameter value

on_epoch_begin(epoch)
set_params(compiled_network, **kwargs)
class ml_genn.callbacks.SpikeRecorder(pop, key=None, example_filter=None, neuron_filter=None, record_counts=False)

Bases: Callback

Callback used for recording spikes during simulation.

Parameters:

  • pop (InputLayer | Layer | Population) – Population to record from

  • example_filter (Sequence | ndarray | integer | int | None) – Filter used to select which examples to record from (see Callbacks and recording for more information).

  • neuron_filter (Sequence | slice | ndarray | integer | int | None) – Filter used to select which neurons to record from (see Callbacks and recording for more information).

  • record_counts (bool) – Should only the (per-neuron) spike count be recorded rather than all the spikes?

get_data()
on_batch_begin(batch)
on_timestep_end(timestep)
set_first()
set_params(data, compiled_network, **kwargs)
class ml_genn.callbacks.VarRecorder(pop, var=None, key=None, example_filter=None, neuron_filter=None, genn_var=None)

Bases: Callback

Callback used for recording state variables during simulation. Variables can specified either by the name of the mlGeNN ml_genn.utils.value.ValueDescriptor class attribute corresponding to the variable e.g. v for the membrane voltage of a ml_genn.neurons.LeakyIntegrateFire neuron or by the internal name of a GeNN state variable e.g. LambdaV which is a state variable added to track gradients by ml_genn.compilers.EventPropCompiler.

Parameters:

  • pop (InputLayer | Layer | Population) – Population to record from

  • var (str | None) – Name of variable to record

  • key – Key to assign recording data produced by this callback in dictionary returned by evaluation/training methods of compiled network

  • example_filter (Sequence | ndarray | integer | int | None) – Filter used to select which examples to record from (see Callbacks and recording for more information).

  • neuron_filter (Sequence | slice | ndarray | integer | int | None) – Filter used to select which neurons to record from (see Callbacks and recording for more information).

  • genn_var (str | None) – Internal name of variable to record

get_data()
on_batch_begin(batch)
Parameters:

batch (int)

on_timestep_end(timestep)
Parameters:

timestep (int)

set_params(data, compiled_network, **kwargs)