quantlab.model.torch_training.cs_rank_norm#

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

Qlib’s CSRankNorm of one bar’s [S_t, L] labels, column by column.

The percentile rank among the finite labels (ties averaged), minus 0.5, times 3.46. NaN stays NaN.

Parameters:
ytorch.Tensor

One bar’s labels, [S_t, L].

Returns:
torch.Tensor

The ranked labels, same shape, device and dtype.

Examples

>>> cs_rank_norm(torch.tensor([[0.3], [float("nan")], [0.1], [0.2]]))
tensor([[ 1.7300],
        [    nan],
        [-0.5767],
        [ 0.5767]])