Re-render feedstock, upgrade version to 1.13.1 #66
Conversation
…da-forge-pinning 2019.02.15
|
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
|
PS: we expect the py37 build to fail because there is no such build for tf 1.12. However, I wanted to start getting the build architecture in place. |
kyleabeauchamp
changed the title
|
It looks like keras_applications is now a pre-req, so I'll have to add conda-forge recipes for that. |
|
keras-applications is the name of the package on conda-forge. |
|
doh, fixed ;) |
|
Hmm, looks like there's an issue where tensorflow fails upon import due to old versions of numpy (from a775aca build), but something is pinning to older versions of numpy. |
|
PS: here is a link to the tf/numpy import issue that I found; issue suggested upgrading version: tensorflow/tensorflow#25942 |
|
I've updated this PR to use the latest 1.13.1 release. I had to disable windows builds to get this to work, but it's now passing all but one of the other platform tests. The remaining Travis CI failure appears to be a timeout (not 100% sure). What do folks think about this? |
kyleabeauchamp
changed the title
kyleabeauchamp
changed the title
|
Yeah given that the osx ones on azure complete (but take 200 minutes) its probably a timeout |
|
in order for this to release binaries we'd need to switch the default publisher |
|
Can you explain that a bit more, never heard of "default publisher". |
| @@ -30,15 +31,17 @@ requirements: | |||
| - absl-py >=0.1.6 | |||
| - astor >=0.6.0 | |||
| - gast >=0.2.0 | |||
| - numpy >=1.13.3 | |||
| - numpy >=1.16.1 | |||
There was a problem hiding this comment.
The wheel indicates compatibility with nuimpy >=1.13.3
There was a problem hiding this comment.
I was unable to get the import to work due to an issue similar to tensorflow/tensorflow#559, which was resolved by a numpy upgrade.
recipe/meta.yaml
Outdated
| @@ -30,15 +31,17 @@ requirements: | |||
| - absl-py >=0.1.6 | |||
| - astor >=0.6.0 | |||
| - gast >=0.2.0 | |||
| - numpy >=1.13.3 | |||
| - numpy >=1.16.1 | |||
| - six >=1.10.0 | |||
| - protobuf >=3.6.0 | |||
recipe/meta.yaml
Outdated
| - termcolor >=1.1.0 | ||
| - grpcio >=1.8.6 | ||
| - enum34 >=1.1.6 # [py<34] | ||
| - backports.weakref >=1.0rc1 # [py<34] | ||
| - mock >=2.0.0 # [py2k] | ||
| - keras-applications |
There was a problem hiding this comment.
has a lower bound of 1.0.6 in the wheel keras-application >=1.0.6
recipe/meta.yaml
Outdated
| - termcolor >=1.1.0 | ||
| - grpcio >=1.8.6 | ||
| - enum34 >=1.1.6 # [py<34] | ||
| - backports.weakref >=1.0rc1 # [py<34] | ||
| - mock >=2.0.0 # [py2k] | ||
| - keras-applications | ||
| - keras-preprocessing |
There was a problem hiding this comment.
Where wheel specifies this as keras-preprocessing >=1.0.5
recipe/meta.yaml
Outdated
| - six >=1.10.0 | ||
| - protobuf >=3.6.0 | ||
| - tensorboard 1.10.* | ||
| #- tensorboard 1.12.* # NB: tensorboard is now bundled in the tensorflow wheel. |
There was a problem hiding this comment.
tensorboard is a separate package on PyPI and in conda. The tensorflow wheel does include the tensorboard entry point but tensorboard is still a dependency.
|
Tensorflow 1.13.1. added a new run dependency of |
|
Made some of the requested changes. |
|
Just need to build some of the other deps: |
Isn't that a terrible version pinning? IIUC that will match all alpha and beta releases of 1.14.0 which surely can't be desired? Even though that version is specified in the tensorflow repo it would probably pay to fix it to |
|
I would be fine with that change, but I'll let @jjhelmus be the deciding vote. |
|
I think we need someone to merge the auto-generated tensorboard bump PR: conda-forge/tensorboard-feedstock#19 |
|
OK the tensorboard dependency is now uploaded, next we need to build the tensorflow-estimator recipe. |
|
OK, made PR for estimator: conda-forge/staged-recipes#7895 |
|
Any progress? |
|
@conda-forge-admin, please rerender |
…da-forge-pinning 2019.04.10
|
This build looks green? Any more reviewers? |
|
LGTM but as @jjhelmus is the assigned reviewer I was going to leave merging to him... |
|
LGTM, thanks for all your work on this @kyleabeauchamp, sorry it too so long for a review. |
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)Wanted to get a recent version of tensorflow built. Once we clean up the recipe and build process here, then we can start planning for the 1.13 release that is upcoming.
FYI I think the tensorflow CPU wheel files use SSE4.2 instructions (and possibly AVX?), so this may push the boundaries of our supported CPU architectures.