getting empty file by Running "feature_extraction_global.py" to extract the global concepts.

[muhammadahsan]

getting empty files by Running "feature_extraction_global.py" to extract the global concepts. The resultant files contains only column names like; ,target,name

[EricFerreiraS]

Hi Muhammad, thanks for reaching out to me. Can you provide more details about your setup? Have you executed the 'Netdissect' approach before executing the code? This step guarantees that the setup is correct. If you have already executed it, can you check your dataset directory and the settings on the 'settings.py' file?

[muhammadahsan]

Yes, I run Netdissect first, then feature_extraction_global.py Actually the below code is returning empty lists of activations_sum, activations_max, activations_avg:

// Function to extract values from the model for each image activations_sum=[] activations_max=[] activations_avg=[] def hook_fn(m, i, o): unit_sum={} unit_max={} unit_avg={} for grad in o:
try: for i,j in enumerate(grad):

sum of all values in the activation map

            unit_sum[i+1]=j.sum().item()
            ##max
            unit_max[i+1]=j.max().item()
            ##avg
            unit_avg[i+1]=j.mean().item()
        activations_sum.append(unit_sum)
        activations_max.append(unit_max)
        activations_avg.append(unit_avg)
    except AttributeError: 
        print ("None found for Gradient")

//Layer that I'm going to use. After all, try to get all layers

layer = model._modules['layer4']

layer.register_forward_hook(hook_fn)

[EricFerreiraS]

There are some points that you have to check:

• What model are you using? The 'layer4' belongs to the ResNet model; if you're using a different model, you need to check the layer's name.
• Check the PyTorch version since the method 'register_forward_hook' should add the function 'hook_fn' into the model.