gatohep.losses#

gatohep.losses.high_bkg_uncertainty_penalty(bkg_sumsq, bkg_yields, rel_threshold=0.2)#

Penalize bins whose relative Monte Carlo uncertainty exceeds a threshold.

Parameters:

  • bkg_sumsq (tf.Tensor) –

    A tensor of shape [ncat] representing the sum of squared weights

    (w_i^2) in each bin.

  • bkg_yields (tf.Tensor) – A tensor of shape [ncat] representing the sum of weights (w_i) in each bin.

  • rel_threshold (float, optional) – The relative uncertainty threshold. Default is 0.2 (20%).

Returns:

A scalar tensor representing the total penalty summed over all bins.

Return type:

tf.Tensor

gatohep.losses.low_bkg_penalty(bkg_yields, threshold=10.0)#

Compute a penalty for background yields below a specified threshold.

Parameters:

  • bkg_yields (tf.Tensor) – A tensor of shape [ncat] representing the background yields in each category.

  • threshold (float, optional) – The minimum background yield threshold. Default is 10.0.

Returns:

A scalar tensor representing the total penalty summed over all categories.

Return type:

tf.Tensor