CrossEntropyLoss Problem for multi-class classification

vision
1

I’m trying to train a classification model: multi-class classification
The loss function used is nn.CrossEntropyLoss()

The batch size is 8 and the number of classes to classify is 6.

The target labels aren’t one-hot encoded - (Checked it)

Screenshot from 2021-11-02 09-43-59
Screenshot from 2021-11-02 09-43-591920×1080 361 KB

This image shows how the target is read from the dataframe and the shape of the target is printed in terminal.

2

The target tensor seems to have an additional dimension and its shape looks like [batch_size, 1]. Remove dim1 for a multi-class classification via label = label.squeeze(1) and it should most likely work.

PS: you can post code snippets by wrapping them into three backticks ```, which would make debugging easier :wink:

1 Like
3

Thank you I figured it out