Log pip freeze (#266)

Log the version of all the python packages active in the python environment (instead of the the version of numpy and pytorch).
This helps in recreating experiments, if you're not using the default versions prescribed in requirements.txt.

distiller/apputils/execution_env.py

@@ -22,11 +22,13 @@ you want to understand the environment in which you execute the training.
 
 import logging
 import logging.config
+import operator
 import os
 import platform
 import shutil
 import sys
 import time
+import pkg_resources
 
 from git import Repo, InvalidGitRepositoryError
 import numpy as np
@@ -83,8 +85,10 @@ def log_execution_env_state(config_paths=None, logdir=None, gitroot='.'):
     if HAVE_LSB:
         logger.debug("OS: %s", lsb_release.get_lsb_information()['DESCRIPTION'])
     logger.debug("Python: %s", sys.version)
-    logger.debug("PyTorch: %s", torch.__version__)
-    logger.debug("Numpy: %s", np.__version__)
+    def _pip_freeze():
+        return {x.key:x.version for x in sorted(pkg_resources.working_set,
+                                                key=operator.attrgetter('key'))}
+    logger.debug("pip freeze: {}".format(_pip_freeze()))
     log_git_state()
     logger.debug("Command line: %s", " ".join(sys.argv))