PymoNNtorch is a Pytorch-adapted version of PymoNNto.
- Free software: MIT license
- Documentation: https://pymonntorch.readthedocs.io.
- Use
torch
tensors and Pytorch-like syntax to create a spiking neural network (SNN). - Simulate an SNN on CPU or GPU.
- Define dynamics of SNN components as
Behavior
modules. - Control over the order of applying different behaviors in each simulation time step.
You can use the same syntax as PymoNNto
to create you network:
from pymonntorch import *
net = Network()
ng = NeuronGroup(net=net, tag="my_neuron", size=100, behavior=None)
SynapseGroup(src=ng, dst=ng, net=net, tag="recurrent_synapse")
net.initialize()
net.simulate_iterations(1000)
Similarly, you can write your own Behavior
Modules with the same logic as PymoNNto
; except using torch
tensors instead of numpy
ndarrays.
from pymonntorch import *
class BasicBehavior(Behavior):
def initialize(self, neurons):
super().initialize(neurons)
neurons.voltage = neurons.vector(mode="zeros")
self.threshold = 1.0
def forward(self, neurons):
firing = neurons.voltage >= self.threshold
neurons.spike = firing.byte()
neurons.voltage[firing] = 0.0 # reset
neurons.voltage *= 0.9 # voltage decay
neurons.voltage += neurons.vector(mode="uniform", density=0.1)
class InputBehavior(Behavior):
def initialize(self, neurons):
super().initialize(neurons)
for synapse in neurons.afferent_synapses['GLUTAMATE']:
synapse.W = synapse.matrix('uniform', density=0.1)
synapse.enabled = synapse.W > 0
def forward(self, neurons):
for synapse in neurons.afferent_synapses['GLUTAMATE']:
neurons.voltage += synapse.W@synapse.src.spike.float() / synapse.src.size * 10
net = Network()
ng = NeuronGroup(net=net,
size=100,
behavior={
1: BasicBehavior(),
2: InputBehavior(),
9: Recorder(['voltage']),
10: EventRecorder(['spike'])
})
SynapseGroup(src=ng, dst=ng, net=net, tag='GLUTAMATE')
net.initialize()
net.simulate_iterations(1000)
import matplotlib.pyplot as plt
plt.plot(net['voltage',0][:, :10])
plt.show()
plt.plot(net['spike.t',0], net['spike.i',0], '.k')
plt.show()
This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.
It changes the codebase of PymoNNto to use torch
rather than numpy
and tensorflow numpy
.