Make edward compatible with Tensorflow 0.12 #374
Conversation
|
All test pass now. Travis will be happy after the release of TF |
|
great work. i approve of all the changes. it's unfortunate that the users have to compile to master get eugene's fix. perhaps it makes sense for now to force all users to use version 0.11; then we wait until 0.12 (non-RC) is available before merging this PR? |
|
Please merge, it is pretty hard to find 0.11 binary, it has been removed from pip. Compiling 0.11 from source is just as hard as installing 0.12rc1 |
|
0.12rc1 does not have the fix, please use these 0.11 binaries for now: https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html#using-pip |
|
Yeah, just figured it out, installed tensorflow from master, then checked out your fork, and then... half of the examples don't work :( |
I have changed a few things in order to make edward compatible with TF 0.12: - `tf.initialize_all_variables()` calls have been renamed to `tf.global_variables_initializer()` - `tf.Variable.ref()` calls are replaced with `value()`. See: tensorflow/tensorflow@c46f70c#diff-ae1a8f7b66539f000615a4ab7e4b2151R460. `ref()` is an internal function now. But I'm not sure whether `value()` is the correct function to use here. - `tf.GraphKeys.VARIABLES` is replaced with `tf.GraphKeys.GLOBAL_VARIABLES`. As a side note, even after applying these changes, edward still doesn't work because of the bug introduced in TF 0.12rc0, see: tensorflow/tensorflow@77cfa97 for the fix. Therefore I had to use TF master branch. There are still 5 test failing, I'm looking into them.
|
I rebased my branch on master to include 1.1.6 hotfix. You can re-clone (not pull) my branch and try again. |
|
thanks! |
|
no news ? Tensorflow-0.12.0 is out since Dec. 20th |
|
This has to be fixed first: tensorflow/tensorflow#5930. |
|
I see they continue to change the API quite a lot since 0.12 |
|
We are rapidly moving toward TensorFlow 1.0. It might make sense to wait a little bit longer. |
dustinvtran
force-pushed
the
master
branch
2 times, most recently
from
a5ac146
to
de5c24c
Compare
|
built off of in #426 |
Initial attempt to add TF 0.12 compatibility.
I have changed a few things in order to make edward compatible with TF 0.12:
tf.initialize_all_variables()calls have been renamed totf.global_variables_initializer()tf.Variable.ref()calls are removed. See: tensorflow/tensorflow@c46f70c#diff-ae1a8f7b66539f000615a4ab7e4b2151R460.ref()is an internal function now.tf.GraphKeys.VARIABLESis replaced withtf.GraphKeys.GLOBAL_VARIABLES.As a side note, even after applying these changes, edward still doesn't work because of the bug introduced in TF 0.12rc0, see: tensorflow/tensorflow#5930 and tensorflow/tensorflow@77cfa97 for the fix. Therefore I had to use TF master branch.
There are still 2 test failing, I'm looking into them.