prune error:dimension out of range (expected to be in range of [-2, 1], but got 3)

[hewumars]

error dimension out of range

values =
torch.sum((activation * grad), dim = 0).
sum(dim=2).sum(dim=3)[0, :, 0, 0].data

[longriyao]

Hi！ How did you solve it?

[Kuldeep-Attri]

Hi!
I am also getting the same error. Did you solve this?
Thanks!!!

[hewumars]

@longriyao @Kuldeep-Attri
pytorch version problem，this repository use pytorch 0.1
If you use pytorch 0.2，add param: keepDim=True

[Kuldeep-Attri]

Thanks!!! @hewumars

[RohitKeshari]

Hi @hewumars
After your suggestion, I have modified and used the following code

values = \
			torch.sum((activation * grad), dim = 0, keepdim=True).\
				sum(dim=2,keepdim=True).sum(dim=3,keepdim=True)[0, :, 0, 0].data

The next error I am getting in prune.py in line 33

RuntimeError: bool value of Variable objects containing non-empty torch.FloatTensor is ambiguous

Before that, I am getting following intermediate output

Accuracy : 0.98454429573
Number of pruning iterations to reduce 67% filters 5
Ranking filters.. 
Layers that will be prunned {0: 1, 2: 5, 5: 6, 7: 6, 10: 21, 12: 16, 14: 21, 17: 64, 19: 64, 21: 60, 24: 62, 26: 79, 28: 107}
Prunning filters.. 

Please suggest how can I fix this problem

[hewumars]

the origin code use python2，if you use 3，please notice the difference between the two versions.

[hewumars]

I do not know where your problem is.

[RohitKeshari]

Python 2.7.13
pytorch 0.2.0_4

I am writing all the massage which I have received while running python finetune.py --prune

/home/iab/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchvision-0.1.9-py2.7.egg/torchvision/transforms/transforms.py:155: UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.
/home/iab/anaconda2/envs/pytorch/lib/python2.7/site-packages/torchvision-0.1.9-py2.7.egg/torchvision/transforms/transforms.py:390: UserWarning: The use of the transforms.RandomSizedCrop transform is deprecated, please use transforms.RandomResizedCrop instead.
Accuracy : 0.98454429573
Number of prunning iterations to reduce 67% filters 5
Ranking filters.. 
Layers that will be prunned {0: 1, 2: 5, 5: 6, 7: 6, 10: 21, 12: 16, 14: 21, 17: 64, 19: 64, 21: 60, 24: 62, 26: 79, 28: 107}
Prunning filters.. 
Traceback (most recent call last):
  File "finetune.py", line 270, in <module>
    fine_tuner.prune()
  File "finetune.py", line 228, in prune
    model = prune_vgg16_conv_layer(model, layer_index, filter_index)
  File "/home/iab/Rohit/pytorch/filter_selection/ICLR2017/pytorch-pruning-master/prune.py", line 33, in prune_vgg16_conv_layer
    bias = conv.bias)
  File "/home/iab/anaconda2/envs/pytorch/lib/python2.7/site-packages/torch/nn/modules/conv.py", line 250, in __init__
    False, _pair(0), groups, bias)
  File "/home/iab/anaconda2/envs/pytorch/lib/python2.7/site-packages/torch/nn/modules/conv.py", line 34, in __init__
    if bias:
  File "/home/iab/anaconda2/envs/pytorch/lib/python2.7/site-packages/torch/autograd/variable.py", line 123, in __bool__
    torch.typename(self.data) + " is ambiguous")
RuntimeError: bool value of Variable objects containing non-empty torch.FloatTensor is ambiguous

Is it because of ambiguous value in Layers that will be prunned?

[hewumars]

@RohitKeshari
#hw modify
is_bias_present = False
if conv.bias is not None:
is_bias_present = True
new_conv =
torch.nn.Conv2d(in_channels = conv.in_channels,
out_channels = conv.out_channels - 1,
kernel_size = conv.kernel_size,
stride = conv.stride,
padding = conv.padding,
dilation = conv.dilation,
groups = conv.groups,
bias = is_bias_present)

[hewumars]

@RohitKeshari step debug and print variable can be solve the problem , because your input param is error.

[RohitKeshari]

Thanks @hewumars, in place conv.bias It should have True. It works for me

[ssxlconch]

Hi,I run the python finetune.py --train
but the output accuracy is always 1.0
and python finetune.py --prune it shows:
Accuracy : 1.0
Number of prunning iterations to reduce 67% filters 5
can you help me ?

[ssxlconch]

@RohitKeshari where did you add the modify code?
I added it:
if not next_conv is None:
next_conv.bias = False
if conv.bias is not None:
next_conv.bias = True
next_new_conv =
torch.nn.Conv2d(in_channels = next_conv.in_channels - 1,
out_channels = next_conv.out_channels,
kernel_size = next_conv.kernel_size,
stride = next_conv.stride,
padding = next_conv.padding,
dilation = next_conv.dilation,
groups = next_conv.groups,
bias = next_conv.bias)
did not solve the problem

[RohitKeshari]

@tearhupo121031 what is your problem? please post the error message here. Are you using the same database? If you are using a different database then it might be the case that database is easy.

[MrLinNing]

sorry,I have the same problem,
I added it:

next_new_conv =
torch.nn.Conv2d(in_channels = next_conv.in_channels - 1,
out_channels = next_conv.out_channels,
kernel_size = next_conv.kernel_size,
stride = next_conv.stride,
padding = next_conv.padding,
dilation = next_conv.dilation,
groups = next_conv.groups,
bias = True)

[CF2220160244]

just delete the last line to this:

if not next_conv is None:
next_conv.bias = False
if conv.bias is not None:
next_conv.bias = True
next_new_conv =
torch.nn.Conv2d(in_channels = next_conv.in_channels - 1,
out_channels = next_conv.out_channels,
kernel_size = next_conv.kernel_size,
stride = next_conv.stride,
padding = next_conv.padding,
dilation = next_conv.dilation,
groups = next_conv.groups)

[ghost]

@MrLinNing

Python 3
model.features._modules.items() -> it returns a generator and you cannot use index

Python 2.7
model.features._modules.items() -> it returns a list and you can use index

You can either use Python 2.7 or list(model.features._modules.items())