Feeding new data to a trained model

1

Hi there

I have trained a conv NN, for 1D (signal) data. my data size is (65535,1,94) which 65535 is number of my samples, 1 is input channel and 94 is length of the signal.

Now I am going to use my trained model for another data set that has the same dimension.
I tried to run this but it did not work:

out_put = torch.zeros(65535,1,94)
for i in range (65535):
out_out[i,1,94] = model (in_put [i,1,94])

below u can also see my network
Autoencoder(
(encoder): Sequential(
(0): Conv1d(1, 5, kernel_size=(5,), stride=(2,))
(1): MaxPool1d(kernel_size=3, stride=1, padding=0, dilation=1, ceil_mode=False)
(2): ReLU(inplace=True)
(3): Conv1d(5, 10, kernel_size=(5,), stride=(2,))
(4): MaxPool1d(kernel_size=3, stride=1, padding=0, dilation=1, ceil_mode=False)
(5): ReLU(inplace=True)
(6): Conv1d(10, 15, kernel_size=(5,), stride=(2,))
(7): MaxPool1d(kernel_size=3, stride=1, padding=0, dilation=1, ceil_mode=False)
(8): ReLU(inplace=True)
(9): Conv1d(15, 20, kernel_size=(4,), stride=(1,))
(10): ReLU(inplace=True)
)
(decoder): Sequential(
(0): ConvTranspose1d(20, 15, kernel_size=(1,), stride=(4,))
(1): ReLU(inplace=True)
(2): ConvTranspose1d(15, 10, kernel_size=(2,), stride=(4,))
(3): ReLU(inplace=True)
(4): ConvTranspose1d(10, 5, kernel_size=(9,), stride=(2,))
(5): ReLU(inplace=True)
(6): ConvTranspose1d(5, 1, kernel_size=(10,), stride=(2,))
(7): ReLU(inplace=True)
)
)

is anyone has any idea about this ?
thank you

2

You would be able to get help if you post the error message as well.

3

Hi

below you can see that

IndexError Traceback (most recent call last)

in () 1 for i in (data_pixel2[0]) : ----> 2 out_put[i,1,94] = model (data_pixel2[i,1,94])

IndexError: index 1 is out of bounds for dimension 1 with size 1

4

Shouldn’t it be out_put[i,0,94] = model (data_pixel2[i,0,94])?

Python is 0-indexed.

1 Like
5

Hi

Actually I got this error again
IndexError Traceback (most recent call last)

in () 1 for i in (data_pixel2[0]) : ----> 2 out_put[i,0,94] = model (data_pixel2[i,0,94])

IndexError: index 94 is out of bounds for dimension 2 with size 94’

6

Python is zero-based as InnovArul already pointed out, index 94 doesn’t exist. That should solve the index problem:
out_put[i] = model(data_pixel2[i]) which is the same as
out_put[i,:,:] = model(data_pixel2[i,:,:])
since you want to input an array of shape (1,94).

In general, I would recommend looking at a pytorch tutorial regarding the structure optimizer and co., e.g. https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html

2 Likes
7

Thank you for replying me back
actually it didn’t work also, I got below error :slight_smile:
RuntimeError Traceback (most recent call last)

in () 1 for i in range(65536) : ----> 2 out_put[i,:,:] = model (data_pixel2[i,:,:])

5 frames

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/conv.py in forward(self, input) 200 _single(0), self.dilation, self.groups) 201 return F.conv1d(input, self.weight, self.bias, self.stride, --> 202 self.padding, self.dilation, self.groups) 203 204

RuntimeError: Expected 3-dimensional input for 3-dimensional weight 5 1 5, but got 2-dimensional input of size [1, 94] instead

8

Oke, the input to the 1D Convolutional layer must be of shape batch size x channel x spatial_size. In your case, it’s only channel x spatial_size. This can be fixed by model(data_pixel2[i].unsqueeze(0)) which adds another dimension in the tensor to be of shape (1,1,94)

9

My data_pixel2 is already (65535,1,94)

here 65536 means number of samples that I do have

10

you’re slicing the first dimension with data_pixel2[i] which leads to a shape of (1,94)

1 Like
11

I see, what should I do now? :frowning:

12

I actually fixed that error, however I now face with this one :
RuntimeError Traceback (most recent call last)

in () 1 for i in range(65536) : ----> 2 out_put[i] = model(data_pixel2[i].unsqueeze(0))

5 frames

/usr/local/lib/python3.6/dist-packages/torch/nn/modules/conv.py in forward(self, input) 200 _single(0), self.dilation, self.groups) 201 return F.conv1d(input, self.weight, self.bias, self.stride, → 202 self.padding, self.dilation, self.groups) 203 204

RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same

13

load your input into VRAM with .cuda()

1 Like
14

I did below
out_put = torch.zeros([65536,1,94],dtype=torch.float)

out_put = out_put.cuda()

still same error …

15

Please have a look into the Pytorch tutorials

16

Could you push data_pixel2 also to the GPU?
The error points to a CPU input tensor, while your model parameters are on the GPU.

1 Like
17

Thanks Patrick
U were right, it fixed :slight_smile:
U are really an expert (Y)

1 Like