Hacker News Reader

Comment by jbverschoor | Hacker News Reader

I have absolutely no idea what this is, supposed to do, and what the demo shows.

To get ML models into production is a lot of work involving tests, retraining, model iteration/management and model serving (often at the edge where latency is low).

ML-Ops is as deep as data engineering but not as often talked about. The linked article discusses one example.

From what I can tell it's an example application to show how a machine learning project could be turned into a real world web app more or less from start to finish.

Yes it covers a bit more than what people tend to see in Machine Learning which is the model training part. From the code base it looks to cover:

1) feature preparation and model training as part of notebooks

2) Creation of a Flask API to interact with the trained models. This includes feature enrichment based on APIs input, so as to match the expected inputs by the model

3) Creation of an UI to interact with the API/Model

4) Setup of NGinx and docker to surface that application

Each part is covered in a minimal manner compared to most enterprise data products, and only cover 1 of the approach for end to end ML, but I think it does the job well to demonstrate the scope of work that is needed to put ML data products into production, and could be used as good introduction.

That's my take as well. yhat, datascience.com, and Domino Data Lab were the first players in this space. Domino has a webinar and trial project to see how these platforms approach this type of workflow. https://www.dominodatalab.com/resources/webinars/web-apps-in...

From the Github: End-to-end machine learning project showing key aspects of developing and deploying real life ml driven application ... a web based ml driven bike trip advisor with trip time prediction

Reply on news.ycombinator.com