numpy / numpy/numpy

BUG: overflow encountered in var() for large float16 arrays

Open
#21,941 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
32.8k
Forks
12.8k
Avg merge
1d 7h
Merged PRs (30d)
197

Description

np.float16(np.random.normal(size=int(1e5))).var()
C:\Users\WIM3SI\.conda\envs\msa\lib\site-packages\numpy\core\_methods.py:247: RuntimeWarning: overflow encountered in reduce
  ret = umr_sum(x, axis, dtype, out, keepdims=keepdims, where=where)
Out[3]: inf

np.float16(np.random.normal(size=int(1e4))).var()
Out[4]: 1.005

np.float32(np.random.normal(size=int(1e5))).var()
Out[5]: 0.9998826

Numpy version: 1.23.0
Python version: 3.9.11
Windows 10

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the float16 and float32 examples, then inspect numpy/core/_methods.py around the var() path and the reported reduce warning. Compare the accumulation behavior for large float16 arrays with the existing tests nearby. Done means the large float16 variance no longer returns inf from overflow while the reported behavior remains covered by a regression test.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
data
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.