PAPIs Latam 2019 has ended
Back To Schedule
Monday, June 24 • 09:00 - 17:00
Building a Simple Recommender System From Scratch — Training Workshop

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
This training workshop will take place before the main conference. It will be given in a small classroom, to maximize interaction and so you can ask even more questions than in a conference setting.

  • A specific ticket is required to get access to the workshop ("Workshop WORKSHOP_NAME + Conference 24-26/6").
  • The venue is different from that of the main conference. The workshop will be held at Dafiti — many thanks to them for providing the space!
  • Students should bring their own laptops, for practical work. 

We will implement and deploy a whole recommender system from scratch, going from simple techniques to more advanced ones. Numpy and Scipy will be used at first, to understand the basic principles; then, a TensorFlow adaptation will be implemented and deployed to production. For this, we will be querying data simulated from Dafiti (which amounts to a few gigabytes), preparing it for the algorithm, training the model serverlessly, and finally building the frontend architecture for the serving step. Google Cloud AI Platform will be used through the whole process (the code and steps that we will present can be used with other public cloud providers, or in private clouds).

Machine learning practitioners interested in learning more about recommender systems and how to fully deploy them into production, going from a real source of data input to recommendations offered to the customers.

  • Basic and advanced Collaborative Filtering algorithms
  • Data processing
  • Deployment of a serverless system following best practices
  • Implement a serverless frontend infrastructure to serve recommendations

Recommendation algorithms:
  • Exploring concepts of collaborative filtering with implicit feedback in Python
  • Going from naive implementations to more sophisticated ones using Numpy / Scipy
  • Adapting the code to TensorFlow (so we can later deploy in a serverless environment)
  • Unit testing

Data preparation:
  • Querying data simulated from Dafiti (~10 GB)
  • Analyzing the data, cleaning it, and preparing it for usage in the recommendation algorithm

  • Preparing the TensorFlow model previously implemented for training on Google Cloud AI Platform
  • Executing a training / test strategy for validating performance

Frontend serving:
  • After our model has been deployed, we then proceed on building a simple Python web service that will handle input requests for recommendations, process the whole thing and send back the processed response
  • A final examination will be made to analyze recommendations performance


avatar for Willian Fuks

Willian Fuks

Data Scientist, Dafiti Group
Working as a Data Scientist at Dafiti Group mainly focused in recommender systems, optimization techniques, A/B testing and search engines. Holds a Master Degree in Artificial Intelligence by Escola Politécnica da USP.

Monday June 24, 2019 09:00 - 17:00 GMT-03
Dafiti Group — Room LYON Av. Francisco Matarazzo, 1350 - Água Branca, São Paulo - SP, 05001-100, Brazil