ANN Regression with TensorFlow on Render.com

Demo

• URL: https://linreg.onrender.com
• Note: Please wait about 2 minutes for the web service to restart if it has been inactive for a long time.
• Then enter: 998 for feature1 and 1000 for feature2, click on "predict" and check the result: 425.12085. More values to check can be found in the fake_reg.csv file.

Jupyter Notebook

• Using Python 3.7.7 on an Intel Mac.
• Eventually just works on Intel PC on CPU, please try

Installation (optional, to regenerate model and scaler)

1. Download and install Miniconda from the website: Miniconda Documentation

2. Create a new environment:

   conda create --name tensorflow377 python=3.7.7
   conda info --envs
   conda activate tensorflow377
   conda install ipykernel
   python -m ipykernel install --user --name tensorflow377 --display-name "Python 3.7.7 (tensorflow)"

3. Load the pip dependencies in Order to retrain model and recreate scaler:

   pip install -r requirements-jpynb.txt

Local Installation of Jupyter Notebook (optional)

1. Deactivate the environment:

   conda deactivate

2. Uninstall Jupyter:

   conda uninstall -y jupyter

3. Install Jupyter and other dependencies:

   conda activate tensorflow377
   pip3 install --upgrade pip
   pip3 install jupyter
   pip install notebook --upgrade
   pip install Jinja2==3.0.3
   pip install MarkupSafe==2.0.0
   pip install zipp==3.1.0
   pip3 install chardet
   conda install -c anaconda importlib-metadata
   conda install -y pandas
   conda install -y seaborn
   conda install -y matplotlib
   conda install -y tensorflow

4. Start the notebook:

   jupyter notebook

Deploying the Application on Render.com

1. Fork this repository on GitHub.
2. Connect the GitHub repository to Render.com (sign in with your GitHub account).
3. Choose the free plan (0$) and the type "Web Service".
4. Build the application on Render.com:

   pip install --upgrade pip && pip install -r requirements.txt

Set Start Command on Render.com

• python app.py

Set Environment Variables on Render.com

• PYTHON_VERSION => 3.7.7
• PORT => 5000

Save Requirements (optional)

• pip freeze > requirements.txt

Scaler (optional)

import pickle

# Save scaler
scalerfile = 'scaler.sav'
pickle.dump(scaler, open(scalerfile, 'wb'))

# Load scaler
scaler = pickle.load(open('scaler.sav', 'rb'))
new_gem2 = [[feat1, feat2]]
new_gem2 = scaler.transform(new_gem2)
predict = model.predict(new_gem2)