Skip to content

switch to using TensorFlow as backend for recent versions of Keras#5821

Merged
wpoely86 merged 2 commits intoeasybuilders:developfrom
boegel:Keras_TensorFlow
Feb 24, 2018
Merged

switch to using TensorFlow as backend for recent versions of Keras#5821
wpoely86 merged 2 commits intoeasybuilders:developfrom
boegel:Keras_TensorFlow

Conversation

@boegel
Copy link
Copy Markdown
Member

@boegel boegel commented Feb 13, 2018 •
edited
Loading

Now that we can properly build TensorFlow from source (cfr. easybuilders/easybuild-easyblocks#1287), we should also use TensorFlow as backend for Keras (which is the default), since it performs a lot better than Theano (which is pretty much end of life).

requires #5819 (TensorFlow)

cc @JensTimmerman

@boegel boegel added the change label Feb 13, 2018
@boegel boegel added this to the next release (3.5.2 or 3.6.0) milestone Feb 13, 2018
@boegel
Copy link
Copy Markdown
Member Author

boegel commented Feb 13, 2018

Test report by @boegel
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in this PR)
node2441.golett.os - Linux centos linux 7.4.1708, Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz, Python 2.7.5
See https://gist.github.com/dad0d3fb966edb15e675e5e7ba3c0943 for a full test report.

@boegel
Copy link
Copy Markdown
Member Author

boegel commented Feb 13, 2018

Test report by @boegel
SUCCESS
Build succeeded for 1 out of 1 (1 easyconfigs in this PR)
node2035.delcatty.os - Linux centos linux 7.4.1708, Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz, Python 2.7.5
See https://gist.github.com/ceccf32166a3ff395ea35e705381bdcf for a full test report.

@boegel boegel mentioned this pull request Feb 13, 2018
Merged
@JensTimmerman
Copy link
Copy Markdown

JensTimmerman commented Feb 13, 2018

On a test application (faceswap) I saw a 5x speed improvement in iteration speed (with similar loss) between a TensorFlow built from source vs theano (built without blas and lapack it warned me) backend.
So yay for tensorflow.

@JensTimmerman
Copy link
Copy Markdown

JensTimmerman commented Feb 14, 2018 •
edited
Loading

I ran into issues with keras 2.1.3 and tensorflow, which did not happen in 2.1.2
Reported upstream, but seems hard to reproduce on their end and hard for me to get a small proof of error.
So yay for tensorflow, but for now only on keras 2.1.2?

@boegel
Copy link
Copy Markdown
Member Author

boegel commented Feb 20, 2018

See issue reported by @JensTimmerman at keras-team/keras#9381 .

Since this is a very clear error, which possible only occurs in same contexts, I don't think that's a blocker for this particular PR...

@easybuilders easybuilders deleted a comment from boegelbot Feb 20, 2018
@boegel boegel requested a review from wpoely86 February 20, 2018 21:58
@wpoely86 wpoely86 merged commit bd72b9b into easybuilders:develop Feb 24, 2018
@boegel boegel deleted the Keras_TensorFlow branch February 25, 2018 11:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wpoely86 wpoely86 wpoely86 approved these changes

Assignees

No one assigned

Labels

change

Projects

None yet

Milestone

3.5.2

Development

Successfully merging this pull request may close these issues.

3 participants

@boegel @JensTimmerman @wpoely86