deep-spin / entmax

The entmax mapping and its loss, a family of sparse softmax alternatives.

deep-spin/entmax Issues

Nans gradients for tensors with singleton dimensions for budget_bisect.
Updated 3 months ago4
Index -1 is out of bounds
Updated 8 months ago1
entmax_bisect bugs with fp16/bf16
Updated 10 months ago1
alpha-entmax
Updated a year ago
EntmaxBisect forward requires unused argument
Closed 2 years ago2
Release patch 1.0.1 with torch install_requires fix
Closed 2 years ago5
Code to reproduce figure 3
Closed 2 years ago1
Entmax fails when all inputs are -inf
Updated 3 years ago1
A bug when alpha = 1 for entmax_bisect?
Closed 3 years ago3
`entmax_bisect` is not stable around `alpha=1`
Closed 3 years ago3
Alpha value less than one?
Closed 3 years ago8
Unexpected behaviour of sparsemax gradients for 3d tensors
Closed 3 years ago8
Problem with sparse activations
Updated 4 years ago9
entmax_bisect leads to loss becoming nan
Closed 4 years ago16
Replicating the behaviour from the paper
Closed 4 years ago1
entmax implementation for Tesnoflow 2
Updated 4 years ago1
Usage of alpha
Closed 5 years ago3
Tensorflow implementation of entmax1.5
Updated 5 years ago10
Errors when using the loss function
Closed 5 years ago9
Sparse losses return nan when there is -inf in the input
Updated 5 years ago2