awslabs/gluonts

`softplus` can return negative values

Open

#1,894 opened on Feb 17, 2022

View on GitHub
 (1 comment) (0 reactions) (0 assignees)Python (3,888 stars) (753 forks)batch import
buggood first issuemxnet

Description

See also fixes needed in #1893 because of this, and issue referenced therein.

To Reproduce

import mxnet as mx
from gluonts.mx.distribution.distribution import softplus
softplus(mx.nd, mx.nd.array([-20]))

Error message or code output

[-1.9073486e-06]
<NDArray 1 @cpu(0)>

Environment

  • Operating system: macos
  • Python version: 3.7.8
  • GluonTS version: a75acdb
  • MXNet version: 1.8.0

Contributor guide

Tech stack
python
Domain
machine learning
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
Python basicsunderstanding of softplus function
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
70
Research direction
Investigate the softplus implementation in gluonts/mx/distribution/distribution.py. The issue shows that softplus returns a small negative value due to numerical precision. Look at the current formula and consider using a numerically stable alternative, such as x > threshold ? x : log(1+exp(x)) or using a built in safe softplus from MXNet. Also check the linked issue #1893 for related fixes and ensure the fix aligns with the broader codebase.