Iterating over Tensor to get samples

vision
1

I’ve a Tensor of shape torch.Size([1291162, 28, 28, 1]). Since this tensor is so big, i decided to take a batch out of it.

yr = x_train[::6400]
print(yr.shape)

This gives back a Tensor of shape

torch.Size([202, 28, 28, 1])

In my train module, i do this

for t in range(2):
    y_pred = model(yr.float())

I want to extract such tensors efficiently in the for loop so that it can be fed to the model. Like I want to grab unique yr like Tensors every time the loop starts and needs to be fed to model accordingly.

2

You could try to use

for _tensor in my_tensor.split(202):
    pred = model(_tensor)
3

I had another doubt. I initialize my model as model = Model(28). My input tensor is [202,1,28,28]. Now Pytorch expects channel size to be placed on the second dim. But I cannot pass to my Model because then it says sizes cannot be non negative.

def conv_layer(ni,nf,kernel_size=3,stride=1):
    return nn.Sequential(
            nn.Conv2d(ni,nf,kernel_size=kernel_size,bias=False,stride=stride,padding=kernel_size//2),
            nn.BatchNorm2d(nf,momentum=0.01),
            nn.LeakyReLU(negative_slope=0.1,inplace=True)
        )

If model(28) is done, then ni takes the value of 28, but then it treats my Tensor as if it has 28 channels, but it has only 1. So how do i pass [202,1,28,28] so that my model treats it as 1 channel image`.