TypeError: softmax() got an unexpected keyword argument 'axis'

[kegl]

On my mac pytest (the MNIST kit) is failing with the error in the subject. I did pip install -U tensorflow. Should we modify the kit?

https://github.com/keras-team/keras/issues/9621

[aboucaud]

Or maybe just pin the dependencies..

[kegl]

I'm not sure pinning is a good idea in this case, we should try to keep the test up with the changes. When I run pytest locally, I don't want to reinstall tensorflow each time.

[aboucaud]

What do you mean ?

This is what virtualenvs are for. So you don't reinstall all the time the needed software.

Le 1 oct. 2018 à 14:55, Balazs Kegl [email protected] a écrit :

I'm not sure pinning is a good idea in this case, we should try to keep the test up with the changes. When I run pytest locally, I don't want to reinstall tensorflow each time.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/paris-saclay-cds/ramp-workflow/issues/150#issuecomment-425896989, or mute the thread https://github.com/notifications/unsubscribe-auth/AC7F3rcRvq6krH-gAXbGqLLpGfShtrKwks5ughDPgaJpZM4XB7jK.

[jorisvandenbossche]

Apart from the discussion about versions, is there actually something to fix? (that we can fix in the code, apart from having the correct version match of keras / tensorflow) And where is the error coming from, from the workflow element in ramp-workflow, or from the example model included in the kit's starting kit submission?

[kegl]

My argument is that the tests in rampwf should be made up to date with the libraries. We can keep the kit itself in a virtualenv, but not rampwf itself.

[kegl]

@jorisvandenbossche the test (pytest) of ramp-workflow is failing on my mac. But I'm not sure if it's because the tensorflow version for mac is not compatible with the latest tensorflow, or the other way around, that the newest tensorflow has this feature.

[aboucaud]

I find some contradictions in the very documentation of tensorflow. They even stipulate that axis is to be preferred as keyword to dim, meaning this one should not be deprecated.

Without axis https://www.tensorflow.org/versions/r1.0/api_docs/python/tf/nn/softmax With axis https://www.tensorflow.org/api_docs/python/tf/nn/softmax

Can you tell us which tensorflow version you are using ?

pip show tensorflow

Le 1 oct. 2018 à 15:03, Balazs Kegl [email protected] a écrit :

@jorisvandenbossche https://github.com/jorisvandenbossche the test (pytest) of ramp-workflow is failing on my mac. But I'm not sure if it's because the tensorflow version for mac is not compatible with the latest tensorflow, or the other way around, that the newest tensorflow has this feature.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/paris-saclay-cds/ramp-workflow/issues/150#issuecomment-425899441, or mute the thread https://github.com/notifications/unsubscribe-auth/AC7F3u4gGb122naBUS9OSkndiWGT1fxPks5ughKGgaJpZM4XB7jK.

[kegl]

silver6:ramp-workflow kegl$ pip install -U tensorflow
Requirement already up-to-date: tensorflow in /Users/kegl/anaconda/lib/python2.7/site-packages (1.1.0)
Requirement already satisfied, skipping upgrade: six>=1.10.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (1.11.0)
Requirement already satisfied, skipping upgrade: werkzeug>=0.11.10 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (0.14.1)
Requirement already satisfied, skipping upgrade: mock>=2.0.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (2.0.0)
Requirement already satisfied, skipping upgrade: numpy>=1.11.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (1.15.2)
Requirement already satisfied, skipping upgrade: protobuf>=3.2.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (3.5.2)
Requirement already satisfied, skipping upgrade: wheel in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (0.29.0)
Requirement already satisfied, skipping upgrade: funcsigs>=1 in /Users/kegl/anaconda/lib/python2.7/site-packages (from mock>=2.0.0->tensorflow) (1.0.2)
Requirement already satisfied, skipping upgrade: pbr>=0.11 in /Users/kegl/anaconda/lib/python2.7/site-packages (from mock>=2.0.0->tensorflow) (4.0.2)
Requirement already satisfied, skipping upgrade: setuptools in /Users/kegl/anaconda/lib/python2.7/site-packages (from protobuf>=3.2.0->tensorflow) (39.0.1)

[aboucaud]

Still does not tell me which tensorflow version you're using.

• we might also think about stopping support for Python 2.

Le 1 oct. 2018 à 15:19, Balazs Kegl [email protected] a écrit :

silver6:ramp-workflow kegl$ pip install -U tensorflow Requirement already up-to-date: tensorflow in /Users/kegl/anaconda/lib/python2.7/site-packages (1.1.0) Requirement already satisfied, skipping upgrade: six>=1.10.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (1.11.0) Requirement already satisfied, skipping upgrade: werkzeug>=0.11.10 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (0.14.1) Requirement already satisfied, skipping upgrade: mock>=2.0.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (2.0.0) Requirement already satisfied, skipping upgrade: numpy>=1.11.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (1.15.2) Requirement already satisfied, skipping upgrade: protobuf>=3.2.0 in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (3.5.2) Requirement already satisfied, skipping upgrade: wheel in /Users/kegl/anaconda/lib/python2.7/site-packages (from tensorflow) (0.29.0) Requirement already satisfied, skipping upgrade: funcsigs>=1 in /Users/kegl/anaconda/lib/python2.7/site-packages (from mock>=2.0.0->tensorflow) (1.0.2) Requirement already satisfied, skipping upgrade: pbr>=0.11 in /Users/kegl/anaconda/lib/python2.7/site-packages (from mock>=2.0.0->tensorflow) (4.0.2) Requirement already satisfied, skipping upgrade: setuptools in /Users/kegl/anaconda/lib/python2.7/site-packages (from protobuf>=3.2.0->tensorflow) (39.0.1) — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/paris-saclay-cds/ramp-workflow/issues/150#issuecomment-425904353, or mute the thread https://github.com/notifications/unsubscribe-auth/AC7F3gFDpZRIarGtD1WktdHB3ZChvimJks5ughZRgaJpZM4XB7jK.

[jorisvandenbossche]

we might also think about stopping support for Python 2.

+1

---

So if I remember correctly, our idea was to keep testing the kits in rampwf-kits-test-master with their specified (pinned) versions, but the embedded sample kits in ramp-workflow itself should be updated to make sure they pass in the tests for all versions of dependencies we want to support in rampwf.

[kegl]

1.1.0

[jorisvandenbossche]

and which keras version?

[kegl]

2.2.0

[kegl]

updated to 2.2.2, same error

[aboucaud]

Ok so current version of tensorflow is 1.11.0 and has the keyword axis. So it's more a local issue.

[jorisvandenbossche]

Is it possible that the combination you have of keras and tensorflow are not compatible?

[kegl]

ok. so no solution, locally.

[aboucaud]

maybe try a newer version of tensorflow, like 1.8 or 1.9

pip install tensorflow==1.9.0

[kegl]

doesn't work

[jorisvandenbossche]

I don't really find any documentation of keras on what versions of tensorflow they support

doesn't work

You mean that it gives the same error?

[kegl]

there is no version for me above 1.1

silver6:ramp-workflow kegl$ pip install tensorflow==1.2.0
Collecting tensorflow==1.2.0
  Could not find a version that satisfies the requirement tensorflow==1.2.0 (from versions: )
No matching distribution found for tensorflow==1.2.0