quanda.utils.functions package

Util funcitons.

quanda.utils.functions.cosine_similarity(test, train, replace_nan=0) Tensor[source]

Compute cosine similarity between test and train activations.

Parameters:
  • test (torch.Tensor) – The test activations.

  • train (torch.Tensor) – The train activations.

  • replace_nan (int, optional) – The value to replace NaN values with. Default is 0.

Returns:

The cosine similarity between the test and train activations.

Return type:

torch.Tensor

quanda.utils.functions.dot_product_similarity(test, train, replace_nan=0) Tensor[source]

Compute cosine similarity between test and train activations.

Parameters:
  • test (torch.Tensor) – The test activations.

  • train (torch.Tensor) – The train activations.

  • replace_nan (int, optional) – The value to replace NaN values with. Default is 0.

Returns:

The dot product similarity between the test and train activations.

Return type:

torch.Tensor

quanda.utils.functions.kendall_rank_corr(tensor1, tensor2)[source]

Calculate torchmetrics kendall_corrcoef function.

The difference is that the input tensors are transposed before passing to the function.

Parameters:
  • tensor1 (torch.Tensor) – The input tensors to compute the correlation coefficient.

  • tensor2 (torch.Tensor) – The input tensors to compute the correlation coefficient.

Returns:

The Kendall Rank correlation coefficient between the two tensors.

Return type:

torch.Tensor

quanda.utils.functions.spearman_rank_corr(tensor1, tensor2)[source]

Calculate torchmetrics spearman_corrcoef function.

The difference is that the input tensors are transposed before passing to the function.

Parameters:
  • tensor1 (torch.Tensor) – The input tensors to compute the correlation coefficient.

  • tensor2 (torch.Tensor) – The input tensors to compute the correlation coefficient.

Returns:

The Spearman correlation coefficient between the two tensors.

Return type:

torch.Tensor

Submodules