Notes

Overview

• With ML applications, the model, scaler, and any other applicable components run in the Python-based backend.

Process

Flask

• Data input through API requests using Postman which relays JSON data to the flask app and Python's request library.
• Flask app connects to HTML web form.

• App
• HTML form using Flask linked to home.html
• Form data input into Flask.
• Flask returns data generated by function through function.html.
• Templates
• home.html
• Flask-based form for accepting user input.
• Relays user input to Flask app.
• function.html
• Returns data generated by funcion.

Heroku Deployment

• In an Anaconda Prompt, navigate to the location of the files to be deployed.
• Create a venv with intended Python version:
Copy

conda create --name venvcustomname python=3.7

• Activate new venv and install relevant libraries for application.
• Use pip, not conda, so that a pip freeze requirements file can be generated.
• Libraries
• flask
• Flask-WTF
• scikit-learn
• This also installs joblib and numpy.
• tensorflow
• gunicorn
• Python web server gateway interface.
• Used to push the flask app to Heroku server.
• Generate pip freeze requirements file to log the installed libraries.
Copy

pip freeze > requirements.txt

Resource: https://pip.pypa.io/en/stable/cli/pip_freeze/

• Create process file required by Heroku.
• Label the file Procfile with no file extension.
• Include the following (note the name of the app file reference):
Copy

web: gunicorn app:app

• On Heroku Dashboard, Create new app.
• Choose unique App name. (my-tf-flower-app)
• Choose Deployment method, namely Heroku Git and follow instructions.
• Push rejected notifications
• No matching distribution found for <libraryname>
• Identify the library that was not able to be installed and identify a suitable version from the available options in Heroku.
• Update the requirements.txt file to reflect the intended version of the library.
• Compile slug size is too large (max is 500M)
• Consider other library versions that may be lighter and without heavy dependencies.
• Resources
• https://devcenter.heroku.com/articles/slug-compiler#slug-size
• https://help.heroku.com/KUFMEES1/my-slug-size-is-too-large-how-can-i-make-it-smaller
• Update the requirements.txt file to reflect the intended version of the library.
• Commit and push update in Prompt
Copy

git commit -am "Updated lib in requirements file."
git push heroku master

• Heroku attempts to compress and build source.
• Successful deployment results in message of
Copy

Launching...
https://<appname>.herokuapp.com/ deployed to Heroku
Verifying deploy... done.

• Deactivate venv in prompt.

Local Testing

Postman

• Use Postman to send a POST request to the API in JSON format.

Troubleshooting

TF on Heroku

• The Tensorflow module is very large, exceeding the slug limits for Heroku, primarily because of the GPU support. To mitigate the slug limits issue that arises, replace tensorflow with tensorflow-cpu in the pip freeze requirements file.

Source: https://stackoverflow.com/questions/61796196/heroku-tensorflow-2-2-1-too-large-for-deployment/62356779#62356779

• Beware that apps on Heroku using TF can take some time to load.

• Notes
• Overview
• Process
• Flask
• Heroku Deployment
• Local Testing
• Postman
• Troubleshooting
• TF on Heroku