Pruning problem

[LKAMING97]

Hello，LightXECG is a good job ,I am very interested in the code of your pruning part. I want to know how to implement it, whether the code can be made public, and if not, can you provide some corresponding reference code?

[lhkhiem28]

Hi @LKAMING97, thanks for your interest in my work
please refer to this code for pruning, note that it should be synchronized with the main repo
https://gist.github.com/lhkhiem28/96494db44ca9278b5cf226190061f7b0

and here are some utility functions

import torch
import torch.nn.utils.prune as prune

def get_pruned_parameters(model, pruned=False):
    total_parameters, total_zero_parameters = 0, 0

    pruned_parameters = []
    for _, module in model.named_modules():
        if isinstance(module, torch.nn.Conv1d) or isinstance(module, torch.nn.Linear):
            if not pruned:
                for _, parameter in module.named_parameters():
                    total_parameters += parameter.nelement()
                    total_zero_parameters += torch.sum(parameter == 0).item()
                    pruned_parameters.append((module, "weight"))
            else:
                for _, parameter in module.named_buffers():
                    total_parameters += parameter.nelement()
                    total_zero_parameters += torch.sum(parameter == 0).item()
                    pruned_parameters.append((module, "weight"))

    return pruned_parameters

def remove_pruned_parameters(model):

    for module_name, module in model.named_modules():
        if isinstance(module, torch.nn.Conv1d):
            try:
                prune.remove(module, "weight")
            except:
                pass
            try:
                prune.remove(module, "bias")
            except:
                pass
        elif isinstance(module, torch.nn.Linear):
            try:
                prune.remove(module, "weight")
            except:
                pass
            try:
                prune.remove(module, "bias")
            except:
                pass
    # for module_name, module in model.named_buffers():
    #     if isinstance(module, torch.nn.Conv1d):
    #         try:
    #             prune.remove(module, "weight")
    #         except:
    #             pass
    #         try:
    #             prune.remove(module, "bias")
    #         except:
    #             pass
    #     elif isinstance(module, torch.nn.Linear):
    #         try:
    #             prune.remove(module, "weight")
    #         except:
    #             pass
    #         try:
    #             prune.remove(module, "bias")
    #         except:
    #             pass

    return model

[LKAMING97]

I would like to ask how the lead-wise grad cam is calculated. Using it, the output of GradCAM in Captum is directly 1 in the channel dimension. How do you deal with it?

[XLHFSJ]

Hello, I read your LightX3ECG paper and was interested in your visualization section or Grad-CAM section, can you publish the code, or can you provide some reference code for it?