Menu
I have a byte tensor of integer class labels, e.g. from the MNIST data set.
1
7
5
[torch.ByteTensor of size 3]
How do use it to create a tensor of 1-hot vectors?
1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 0 0 0
0 0 0 0 1 0 0 0 0 0
[torch.DoubleTensor of size 3x10]
I know I could do this with a loop, but I'm wondering if there's any clever Torch indexing that will get it for me in a single line.
694
In the above example, we try to implement the one hot() encoding function as shown here first; we import all required packages such as a torch. After that, we created a tenor by using the rand() function as shown, and finally, we applied the one hot() function with the argmax() function as shown.
To create a tensor with pre-existing data, use torch.tensor() . To create a tensor with specific size, use torch.* tensor creation ops (see Creation Ops). To create a tensor with the same size (and similar types) as another tensor, use torch.*_like tensor creation ops (see Creation Ops).
public class TensorLabel. TensorLabel is an util wrapper for TensorBuffers with meaningful labels on an axis. For example, an image classification model may have an output tensor with shape as {1, 10}, where 1 is the batch size and 10 is the number of categories.
FloatTensor ; by default, PyTorch tensors are populated with 32-bit floating point numbers.
An alternate method is to shuffle rows from an identity matrix:
indicies = torch.LongTensor{1,7,5}
one_hot = torch.eye(10):index(1, indicies)
This was not my idea, I found it in karpathy/char-rnn.
98
indices = torch.LongTensor{1,7,5}:view(-1,1)
one_hot = torch.zeros(3, 10)
one_hot:scatter(2, indices, 1)
You can find the documentation for scatter in the torch/torch7 github readme (in the master branch).
25
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
