make tensorpack work with tf eager

[eyaler]

i tried running some examples with tfe but could not get it to work
https://research.googleblog.com/2017/10/eager-execution-imperative-define-by.html

[ppwwyyxx]

One big distinction between eager mode and graph mode:

In eager mode, variable creation and the model math must be separate. Variables are created once, and the model is then executed many times. However, in graph mode they are both run only once and therefore they are usually together. A tf.layers.dense(x) call will create both the variables and the math operations.

As a result, in eager mode you'll almost always have to rewrite your model code anyway, into pytorch-style or Keras-style. That's not a tensorpack limitation. Eager documentation also mentions:

Some API calls (such as the functional-style tf.layers.dense, tf.layers.conv2d) are not compatible with eager execution. Use of such methods should raise an error indicating the alternative (e.g., the tf.layers.Dense and tf.layers.Conv2D classes).

So probably this feature won't happen soon.