Add wip notebook

Author: Dan Choi

advanced_functionality/tensorflow_bring_your_own/Dockerfile renamed to advanced_functionality/tensorflow_bring_your_own/container/Dockerfile

@@ -10,5 +10,5 @@ RUN apt-get update && apt-get install tensorflow-model-server
 
 ENV PATH="/opt/ml/code:${PATH}"
 
-COPY src /opt/ml/code
+COPY /cifar10 /opt/ml/code
 WORKDIR /opt/ml/code

advanced_functionality/tensorflow_bring_your_own/src/__init__.py renamed to advanced_functionality/tensorflow_bring_your_own/container/cifar10/__init__.py

@@ -16,7 +16,6 @@
 
 import argparse
 import functools
-import logging
 
 import os
 
@@ -47,8 +46,6 @@
 
 _BATCHES_PER_EPOCH = NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN / BATCH_SIZE
 
-logging.basicConfig(format='%(asctime)s %(name)-12s %(levelname)-8s %(message)s', level=logging.DEBUG)
-
 
 def model_fn(features, labels, mode):
     """Model function for CIFAR-10."""
@@ -186,7 +183,6 @@ def train(model_dir, data_dir, train_steps):
 def main(model_dir, data_dir, train_steps):
     tf.logging.set_verbosity(tf.logging.INFO)
     train(model_dir, data_dir, train_steps)
-    print('Training Done!')
 
 
 if __name__ == '__main__':

advanced_functionality/tensorflow_bring_your_own/src/cifar10.py renamed to advanced_functionality/tensorflow_bring_your_own/container/cifar10/cifar10.py

@@ -8,14 +8,16 @@ events {
 http {
   server {
     # configures the server to listen to the port 8080
+    # https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-inference-code.html#your-algorithms-inference-code-container-response
     listen 8080 deferred;
 
     # redirects requests from SageMaker to TF Serving
     location /invocations {
       proxy_pass http://localhost:8501/v1/models/cifar10_model:predict;
     }
 
-    # Used my SageMaker to confirm if server is alive.
+    # Used by SageMaker to confirm if server is alive.
+    # https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-inference-code.html#your-algorithms-inference-algo-ping-requests
     location /ping {
       return 200 "OK";
     }

advanced_functionality/tensorflow_bring_your_own/src/nginx.conf renamed to advanced_functionality/tensorflow_bring_your_own/container/cifar10/nginx.conf

@@ -0,0 +1,26 @@
+#!/usr/bin/env python
+
+# This file implements the hosting solution, which just starts TensorFlow Model Serving.
+
+from __future__ import print_function
+
+import subprocess
+
+
+def start_server():
+    print('Starting TensorFlow Serving.')
+
+    # link the log streams to stdout/err so they will be logged to the container logs
+    subprocess.check_call(['ln', '-sf', '/dev/stdout', '/var/log/nginx/access.log'])
+    subprocess.check_call(['ln', '-sf', '/dev/stderr', '/var/log/nginx/error.log'])
+
+    nginx = subprocess.Popen(['nginx', '-c', '/opt/ml/code/nginx.conf'])
+    tf_model_server = subprocess.call(['tensorflow_model_server',
+                                       '--rest_api_port=8501',
+                                       '--model_name=cifar10_model',
+                                       '--model_base_path=/opt/ml/model/export/Servo'])
+
+
+# The main routine just invokes the start function.
+if __name__ == '__main__':
+    start_server()

advanced_functionality/tensorflow_bring_your_own/src/resnet_model.py renamed to advanced_functionality/tensorflow_bring_your_own/container/cifar10/resnet_model.py

@@ -6,19 +6,15 @@
 
 from __future__ import print_function
 
-import logging
 import os
 import json
 import sys
 import subprocess
 import traceback
 
 # These are the paths to where SageMaker mounts interesting things in your container.
-from fcntl import fcntl, F_GETFL, F_SETFL
-
 prefix = '/opt/ml/'
-
-input_path = prefix + 'input/data'
+input_path = os.path.join(prefix,'input/data')
 output_path = os.path.join(prefix, 'output')
 model_path = os.path.join(prefix, 'model')
 param_path = os.path.join(prefix, 'input/config/hyperparameters.json')
@@ -32,49 +28,10 @@ training_path = os.path.join(input_path, channel_name)
 training_script = 'cifar10.py'
 default_params = ['--model-dir', str(model_path)]
 
-logging.basicConfig(format='%(asctime)s %(name)-12s %(levelname)-8s %(message)s', level=logging.DEBUG)
-
-
-def _stream_output(process):
-    """Stream the output of a process to stdout
-    This function takes an existing process that will be polled for output. Both stdout and
-    stderr will be polled and both will be sent to sys.stdout.
-    Args:
-        process(subprocess.Popen): a process that has been started with
-            stdout=PIPE and stderr=PIPE
-    Returns (int): process exit code
-    """
-    exit_code = None
-
-    # Get the current flags for the  stderr file descriptor
-    # And add the NONBLOCK flag to allow us to read even if there is no data.
-    # Since usually stderr will be empty unless there is an error.
-    flags = fcntl(process.stderr, F_GETFL)  # get current process.stderr flags
-    fcntl(process.stderr, F_SETFL, flags | os.O_NONBLOCK)
-
-    while exit_code is None:
-        stdout = process.stdout.readline().decode("utf-8")
-        sys.stdout.write(stdout)
-        try:
-            stderr = process.stderr.readline().decode("utf-8")
-            sys.stdout.write(stderr)
-        except IOError:
-            # If there is nothing to read on stderr we will get an IOError
-            # this is fine.
-            pass
-
-        exit_code = process.poll()
-
-    if exit_code != 0:
-        raise RuntimeError("Process exited with code: %s" % exit_code)
-
-    return exit_code
-
 
-# The function to execute the training.
+# Execute your training algorithm.
 def _run(cmd):
     process = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, env=os.environ)
-    _stream_output(process)
     stdout, stderr = process.communicate()
 
     return_code = process.poll()
@@ -102,7 +59,6 @@ if __name__ == '__main__':
         cmd_args = _hyperparameters_to_cmd_args(training_params)
 
         train_cmd = [python_executable, training_script] + default_params + cmd_args
-        print(train_cmd)
 
         _run(train_cmd)
         print('Training complete.')