Allow NN ensemble to be used for parallel eval #515

osma · 2021-08-17T13:16:02Z

Fixes #453 by introducing a parallel parameter to project.initialize() and backend.initialize() methods. Setting it to True warns the project/backend that it will be used for parallel execution. The NN ensemble heeds this warning and delays loading the TensorFlow model until later, avoiding the lockup that otherwise happens.

Some unscientific benchmark results for the eval command examples in #453 using i5-7200U:

-	user time (s)	wall time (m:s)
Before PR	279	4:42
jobs=1	251	4:11
jobs=2	329	2:47
jobs=0	413	2:04

Opening draft PR for QA tool results.

…453

codecov · 2021-08-17T13:16:11Z

Codecov Report

Merging #515 (e47d1c7) into master (845985a) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #515      +/-   ##
==========================================
+ Coverage   99.51%   99.53%   +0.01%     
==========================================
  Files          82       82              
  Lines        5822     5831       +9     
==========================================
+ Hits         5794     5804      +10     
+ Misses         28       27       -1

Impacted Files	Coverage Δ
annif/backend/backend.py	`100.00% <100.00%> (ø)`
annif/backend/dummy.py	`100.00% <100.00%> (ø)`
annif/backend/ensemble.py	`100.00% <100.00%> (ø)`
annif/backend/fasttext.py	`97.82% <100.00%> (ø)`
annif/backend/mllm.py	`100.00% <100.00%> (ø)`
annif/backend/nn_ensemble.py	`99.25% <100.00%> (+0.01%)`	⬆️
annif/backend/omikuji.py	`98.76% <100.00%> (ø)`
annif/backend/pav.py	`98.90% <100.00%> (ø)`
annif/backend/stwfsa.py	`100.00% <100.00%> (+1.56%)`	⬆️
annif/backend/svc.py	`100.00% <100.00%> (ø)`
... and 7 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 845985a...e47d1c7. Read the comment docs.