quantlab.model.torch_training.cs_zscore#

quantlab.model.torch_training.cs_zscore(y)[source]#

Z-score of one bar’s [S_t, L] labels, column by column.

Subtracts the mean of the finite labels and divides by their sample standard deviation. NaN stays NaN; a column with fewer than two finite labels, or no spread, becomes NaN.

Parameters:

ytorch.Tensor: One bar’s labels, [S_t, L].

Returns:

torch.Tensor: The standardised labels, same shape.

Examples

>>> cs_zscore(torch.tensor([[1.0], [3.0], [float("nan")]]))
tensor([[-0.7071],
        [ 0.7071],
        [    nan]])