Floating point precision error

[zechengtang]

There is a floating point precision error when using 0.02 at each time step. This issue can be reproduced with the following code:

from spikingjelly.activation_based import neuron
import torch

if_layer = neuron.IFNode(step_mode='s')

print(' -'* 6)
if_layer.reset()
print(f"init v: {if_layer.v}")
x = torch.FloatTensor([0.05])
print(f"input {x}")
print(' -'* 6)
for i in range(100):
    out = if_layer(x)
    if out == 1:
        print(f"if_layer fires at step {i}")
        
print(' -'* 6)
if_layer.reset()
print(f"init v: {if_layer.v}")
x = torch.FloatTensor([0.02])
print(f"input {x}")
print(' -'* 6)

for i in range(200):
    out = if_layer(x)
    if out == 1:
        print(f"if_layer fires at step {i}")
        
print(' -'* 6)
if_layer.reset()
print(f"init v: {if_layer.v}")
x = torch.FloatTensor([0.02])
print(f"input {x}")
print(' -'* 6)

for i in range(50):
    out = if_layer(x)
print(f"membrane at step 50:   {if_layer.v}")
print(f"membrane - 1.0:        {if_layer.v - 1.0}")

It appears that the error stems from float32 precision limitations rather than any specific code within this project. I recommend adding a parameter eps with a default value of 1e-6 and changing the judging condition from v >= v_threshold to v + eps >= v_threshold. Or you could just adjust the variable x used in your tutorials (https://spikingjelly.readthedocs.io/zh-cn/0.0.0.0.14/activation_based/neuron.html#id4) to avoid any misunderstanding.