TensorFlow Cloud Foundry App

This is a sample app showing how to serve a TensorFlow model in a Python Cloud Foundry app. This uses the Cloud Foundry offering provided by IBM Cloud. The model file and labels file should be put in the assets directory. Currently the files there correspond to a simple fruit classifier which was created by retraining MobileNet_v1_1.0_224.

1. Clone the sample app

Now you're ready to start working with the app. Clone the repo and change to the directory where the sample app is located.

git clone https://github.com/pvaneck/tensorflow-cf-app
cd tensorflow-cf-app

2. Run the app locally

Install the dependencies listed in the requirements.txt file to be able to run the app locally.

You can optionally use a virtual environment to avoid having these dependencies clash with those of other Python projects or your operating system.

pip install -r requirements.txt

Run the app.

python app.py

View your app at: http://localhost:8000

3. Prepare the app for deployment

To deploy to IBM Cloud, it can be helpful to set up a manifest.yml file. One is provided here. Take a moment to look at it.

The manifest.yml includes basic information about your app, such as the name, how much memory to allocate for each instance and the route. In this manifest.yml random-route: true generates a random route for your app to prevent your route from colliding with others. You can replace random-route: true with host: myChosenHostName, supplying a host name of your choice. Learn more...

Also, the amount of memory your app uses is dependent on the size of your model. Feel free to increase or decrease it as you see fit.

applications:
- name: TF-Demo
  random-route: true
  memory: 1024M

4. Deploy the app

You can use the Cloud Foundry CLI to deploy apps.

Choose your API endpoint

cf api <API-endpoint>

Replace the API-endpoint in the command with an API endpoint from the following list.

URL	Region
https://api.ng.bluemix.net	US South
https://api.eu-de.bluemix.net	Germany
https://api.eu-gb.bluemix.net	United Kingdom
https://api.au-syd.bluemix.net	Sydney

cf login

From within the tensorflow-cf-app directory push your app to IBM Cloud.

cf push

This can take a minute. If there is an error in the deployment process you can use the command cf logs <Your-App-Name> --recent to troubleshoot.

When deployment completes you should see a message indicating that your app is running. View your app at the URL listed in the output of the push command. You can also issue the

cf apps

command to view your apps status and see the URL.