LIF Process models fail/hang with use_graded_spike=True

[epaxon]

Objective of issue: Fix LIF ProcessModels so that graded spike work.

Lava version:

• [ ] 0.5.0 (feature release)
• [ ] 0.4.1 (bug fixes)
• [x] 0.4.0 (current version)
• [ ] 0.3.0
• [ ] 0.1.2

I'm submitting a ...

• [x] bug report
• [ ] feature request
• [ ] documentation request

Current behavior:

• using the LIF process with use_graded_spike=True fails, system crashes or hangs at runtime

Expected behavior:

• not failing...

Steps to reproduce:

from lava.proc.io import sink, source from lava.proc.lif.process import LIF from lava.proc.dense.process import Dense

from lava.magma.core.run_configs import Loihi1SimCfg from lava.magma.core.run_conditions import RunSteps vec = np.array([5, 100, 0, 20]) weights = np.zeros((3,4)) weights[:, 0] = [10, 5, 0] weights[:, 1] = [5, 5, 10] weights[:, 2] = [0, 10, 5] weights[:, 3] = [5, 50, 5]

inport_plug = source.RingBuffer(data=vec) dense_weights = Dense(shape=weights.shape, weights=weights, use_graded_spike=True) lif_layer = LIF(shape=(3,), vth=10, du=1, dv=1, use_graded_spike=True)

outport_plug = sink.RingBuffer(shape=(3,), buffer=10)

inport_plug.s_out.connect(dense_weights.s_in) dense_weights.a_out.connect(lif_layer.a_in) lif_layer.s_out.connect(outport_plug.a_in)

lif_layer.run(condition=RunSteps(num_steps=3), run_cfg=Loihi1SimCfg())

Related code:

I have fixed this. There are two problems in the LIF Process model. First the variable "vth_effective" is not defined at the process, but this variable is not needed and can be replaced with "vth". Second, the function reset_voltage is failing when use_grarded_spike=True because the variable "spike_vector" is no longer a boolean numpy array. This can be fixed by changing reset_voltage to force spike_vector to be boolean, for instance:

def reset_voltage(self, spike_vector: np.ndarray):
    """Voltage reset behaviour. This can differ for different neuron
    models."""
    self.v[spike_vector > 0] = 0
    # perhaps like this is more appropriate:
    # self.v[spike_vector ~= 0] = 0

insert short code snippets here

Other information:

insert the output from lava debug here