aws-dlami-tensorflow

This demo illustrates working with TensorFlow using an Amazon Deep Learning AMI (DLAMI). It includes:

1. src/basics.py - Understand basic operations in TensorFlow
2. src/nn.py - Create a small neural network regression model in TensorFlow
3. src/nn_with_summaries.py - Show how to augment code with TensorFlow summaries to visualize the graph and learning process in TensorBoard
4. src/nn_export.py - Illustrates how to save a TensorFlow model to disk so that it can be served by TensorFlow Serving
5. src/nn_client.py - Example of how to consume the model served by TensorFlow Serving

Getting Started

Deploy the CloudFormation stack in the template in infrastructure/. The template creates a user with the following credentials and minimal required permisisons to complete the Lab:

• Username: student
• Password: password

Instructions

• Connect to the instance using the SSH username: ubuntu.
• Run the Jupyter notebook server that comes pre-installed on the Amazon Deep Learning AMI: jupyter notebook
• SSH tunnel to the notebook server running on port 8888
• Open a browser to the notebook server on localhost. Get the URL with token from the command jupyter notebook list
• Create a new python 2.7 and TensorFlow environment notebook for each file in the src/ directory
• Paste the code in from each script in the src/ directory into a cell
• Run the notebooks
  • To view the summaries of src/nn_with_summaries.py in TensorBoard, run the command: tensorboard --logdir /tmp/tensorflow/nn
  • To serve the model saved by src/nn_export.py with TensorFlow Serving, run the command: tensorflow_model_server --port=9000 --model_name=nn --model_base_path=/tmp/nn

Cleaning Up

Delete the CloudFormation stack to remove all the resources. No resources are created outside of those created by the template.