Port Coursera to PyTorch

[Dmitry Nikulin]

Tensorflow is hard to debug and hard to teach with. We are rolling out an experimental course version using PyTorch.

The following files will have to be ported:

• week1_intro/primer/recap_tensorflow.ipynb
• week4_approx/dqn_atari.ipynb
• week4_approx/framebuffer.py
• week4_approx/practice_approx_qlearning.ipynb
• week5_policy_based/atari_util.py
• week5_policy_based/practice_a3c.ipynb
• week5_policy_based/practice_reinforce.ipynb
• week6_outro/seq2seq/basic_model_tf.py
• week6_outro/seq2seq/practice_tf.ipynb

PyTorch notebooks will initially be released side-by-side with TF notebooks (and will be connected to the same graders). We will then consider removing TF entirely.

[Dmitry Nikulin]

Almost fully fixed in #343. Will create a separate issue for week5_policy_based/practice_a3c.ipynb