TensorFlow compatibility problems

[TITAN-PyCompat]

Dear developers,

I observed some TensorFlow api usage in you code that might lead to serious compatibility problems(like crash).

In api tf.cond, the parameter true_fn and false_fn appeared in tf v1.2.0-rc0 and not existed in earlier version. The corresponding parameter is fn1 and fn2 in older version. If you use the new version parameter name to pass the value, it will crash in earlier version.

Same problem happens to api tf.concat, parameter axis appears in tf v1.0.0. The older parameter is called concat_dim in earlier version api. Crash also happens for earlier version.

These problems can be fixed in pr #66 I judged the version and used parameter name to make it compatible on more versions.

Look forward to your reply! @yangJirui @yangxue0827 Thanks!

[yangxue0827]

Thank you very much for your support of our project. We also found the problem you reflected, but the high-frequency use of these functions throughout the project, the solution you provide will make the whole project appear more bloated, so we recommend using a higher version of tensorflow instead of doing so. Of course there may be better ways to solve this problem, such as encapsulating these functions. Thanks again. @TITAN-PyCompat

[TITAN-PyCompat]

@yangxue0827 Thanks for the reply. However, I just create another pull request to fix another kind of compatibility problem inducing API usage.

For tf.global_variables_initializer,tf.local_variables_initializer,tf.summary.image appears in tf v0.12.0-rc0 and this will induced a crash if used in earlier version. Problems happen also to tf.summary.scalar which appeared in tf v0.11.0rc0.

Before seeing this reply, I just made some patches for your code in another pr #68 . I agree with your reply. There is some better way to do this. So maybe you won't merge my code, but the problems are real, right?

[yangxue0827]

Why do you still insist on using a lower version of tf? @TITAN-PyCompat