Mini project; Following this blog post step by step to make a XGBoost model created through Scikit-learn available through a Lambda function using Zappa.
Nice tutorial and easy to follow, though I’m unsure how the load_model() function could have worked when trying to create the already existing bucket every time it’s called, which for me throws an error (…and thus changed to conn.get_bucket(BUCKET_NAME) instead). Additionally, I changed out the json encoding function to return pd.Series(prediction[-1]).to_json(orient='index') to get it to work with my model.
Further, the tutorial is almost a year old, and Zappa doesn’t work with pip 10.0.1 currently, so I had to downgrade to 9.0.3. This is apparently a known issue.
Now, the API and model works beautiful locally and it should just be the zappa deploy left to go, and how sweet that would be. As the project tarball ends up being closer to 90mbs though, trying to deploy just ends up with timeouts from the command line though I have quite decent upload speed. Trying to change the Boto timeout does nothing to help.
After trying for an hour to find a solution, I give up.
Status: Fail
Comments