Deployment of Flask App with Iris KI (ANN)

• Udemy Course: Deep Learning with TensorFlow - Chapter 20: Deployment
• Demo: https://iris-ann.onrender.com
• Please wait for approx. 2 min. until the queued webservice (due to long inactivity) is restarted.

Login with Google to Render.com Account

• Render Dashboard
• Better login via GitHub.com so you can deploy by using this (e.g. forked) repo.
• For deployment, it's easiest to use a GitHub repository.
• Choose the free plan with 512 MB and 0.1 CPU.

Install Miniconda if Running Locally

• Miniconda Installation Guide

Windows, Linux, or macOS x64 Intel (IMPORTANT: No M1 Mac supported with Python version 3.7.0)

conda create --name tensorflow37 python=3.7.0
conda activate tensorflow37

Installation of dependencies locally (optional)

• pip install gunicorn

• pip install flask

• pip install Flask-WTF

• pip install tensorflow

• pip install scikit-learn

• pip install numpy

• pip freeze > requirements.txt # dont do this

• Optional:

• pip show <PACKAGE> check versions of package

• Content of requirements.txt, don't neet to regenerate:

Flask==2.0.3
Flask-WTF==0.15.1
gunicorn==20.1.0
joblib==1.1.1
scikit-learn==0.21.3
scipy==1.5.2
tensorflow==2.6.2
werkzeug==2.0.3

run locally

• conda activate tensorflow37

build (do this to install dependencies)

• pip install -r requirements.txt

Set Env Vars in render.com choose app then choose Environment menu item

• Only thing needed for sure is: PYTHON_VERSION / 3.7.7
• and maybe:
• PORT / 5000
• optional:
• PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION / python

Start command

• replace: gunicorn app:app
• dont use: gunicorn --timeout 600 --bind 0.0.0.0:$PORT app:app # timeout error
• instead use:
• python app.py #also in render backend configuration

generation of saved model and saved scaler see:

• https://github.com/peterruler/iris-ann-ipynb