Upgrade to TensorFlow 1.13.1 (aws#184)

* Upgrade to TensorFlow 1.13.1

* Add 1.13.1 dockerfiles
* Lower process number for one of the horovod local mode test
    Too many processes running in parallel causes the test fail sometimes.
* Use multiprocessing.Process to start ps

Author: icywang86rui

docker/1.13.1/Dockerfile.cpu

@@ -107,4 +107,4 @@ RUN ${PIP} install --no-cache-dir -U \
 
 ENV SAGEMAKER_TRAINING_MODULE sagemaker_tensorflow_container.training:main
 
-CMD ["bin/bash"]
+CMD ["bin/bash"]

src/sagemaker_tensorflow_container/training.py

@@ -17,7 +17,6 @@
 import multiprocessing
 import os
 import subprocess
-import threading
 import time
 
 import sagemaker_containers.beta.framework as framework