Project Eli

This is a final undergraduate (level 4) project by Matej Babis.

Languages: Python, Javascript & HTML

Webhook: https://projecteli.ngrok.io/

• Spotify Client ID: N/A (edited)
• Dialogflow Client ID: N/A (edited)
• ngrok Pro token: N/A (edited)

Requirements:

• GPy
• SciPy, NumPy, matplotlib
• Flask
• ngrok
  • Dialogflow API requires a webhook to be connected to. By running a Flask Python app on localhost, we can tunnel the domain onto a local machine.
  • In order to simplify the installation process, a Pro version of ngrok has been purchased.
    • Add ngrok to the project directory.
    • Add the paid token: ./ngrok authtoken TOKEN.

After installing the dependencies, do the following:

1. Add the following function to {YOUR_PYTHON_ROOT}/Lib/site-packages/GPy/core/gp.py:

def predict_f(self, _Xnew, full_cov=False, kern=None):
   mu, var = self._raw_predict(_Xnew, full_cov=full_cov, kern=kern)
   return mu, var

2. Add the following line to /Lib/site-packages/GPy/kern/__init__.py:

from .src.pjk import PjkRbf

3. Copy the pjk.py file into /Lib/site-packages/GPy/kern/src/.
   • This is the custom covariance function.

Execute by:

> python app.py
> ./ngrok http -subdomain=projecteli 5000

Now you can access https://projecteli.ngrok.io/.

Alternatively, you can run the model in a shell without a front-end by executing:

> python offline.py

Bug list

• Occasinally no response is produced for a command.
  • Response is sent from the server, so possibly a Javascript issue.
  • Mitigation: Use the command again.