unchanging minibatches from skflow DataFeeder for unsupervised tasks #2159
Comments
|
Agreed that this looks like a bug. Thanks for the find! Feel free to submit a pull request (I can also submit a fix :) |
|
You're welcome. I just happened to come across it when looking at how the data feeders work. I don't think I'll have bandwidth for a PR in the near future, but I thought it'd be helpful to make note of it here. |
|
Thanks again! I'll submit the fix :) |
vrv
pushed a commit
to vrv/tensorflow
that referenced
this issue
May 5, 2016
…man. tensorflow#2159 Change: 121133289
|
@cssndrx: Did this fix make it into the master branch? |
|
Great, thanks! |
qbx2
pushed a commit
to qbx2/tensorflow
that referenced
this issue
May 25, 2016
…man. tensorflow#2159 Change: 121133289
fsx950223
pushed a commit
to fsx950223/tensorflow
that referenced
this issue
Dec 22, 2023
…builds2 fix autobuilds
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I'm pretty sure there's a bug here in the DataFeeder class for skflow (tf.contrib.learn). If there are no labels, then the mapping of input placeholders to inputs is returned without updating the minibatch offsets. The code for updating offsets and shuffling won't be reached.
The text was updated successfully, but these errors were encountered: