Albert Zeyer

@Spotlight0xff very interesting. I'm not sure if this is the same hang though. [This TF issue (deadlock in `fork()` because of OpenBlas)](https://github.com/tensorflow/tensorflow/issues/13802) sounds very related. (I wonder now why I...

To debug this, it would be helpful if we somehow can investigate what the GPU is doing at that time (similarly as we are seeing what the CPU is doing)....

@JackTemaki I heard you also got this? You did not comment here. @Marvin84 You as well? You also did not comment here. (It would have been useful to know that...

All of you always had this problem only when 2D convolution was involved, in the typical VGG-style prenet we have now in many models?

Also, you always get this only after min 1 day of runtime? Or even longer, more like 2.5 days? Is there some common pattern? Or is it more related to...

All the reports here are also for training, right?

@curufinwe pointed out that `tf_session_opts = {'inter_op_parallelism_threads' : 1}` can potentially avoid the problem. This is obviously not a real solution. Maybe it just hides the problem. But it is...

This here is a case where we do not necessarily need a new `behavior_version` and instead could just print a deprecation warning.

> There you mentioned that the global naming is not really nice right now. Would it make sense to maybe give them a prefix similar to data (like `target:`) in...

They are not exactly redundant or the same. `learning_rate` defines the initial and default learning rate. `learing_rates` sets predefined learning rates for some specified epochs.