chris j wallace

Chris J Wallace

Hi, I'm Chris. I'm a data scientist, hacker and person who cares. I like building data products and making technology work for people, not vice versa. I was a particle physicist once. I climb small walls and play loud guitars. I don't  tweet  much.

Blog

Personal posts

Speaking is Hard

March 28, 2019

Writing is Hard

February 11, 2019

Fast Forward Labs posts

NBSVM: a strong document (topic) classification baseline

August 14, 2019

Engineering for reproducibility

June 03, 2019

AutoML, open and closed

April 29, 2019

TensorFlow Dev Summit 2019

April 08, 2019

Neural Ordinary Differential Equations

March 07, 2019

Projects

Fizzing the proverbial buzz.

NBSVM

A custom scikit-learn classifier implementing the NBSVM model described in Baselines and Bigrams: Simple, Good Sentiment and Topic Classification and used as the baseline for the Fast Forward Labs report on transfer learning for NLP.

An Invitation to Active Learning

Observable notebook in which I construct a demonstration of active learning with logistic regression live in the browser with TensorFlow.js. Required some fiddling with the reactive cell evaluation of Observable.

Shakespair

Observable notebook in which a recurrent neural net trained on the corpus of Shakespeare plays completes your sentence, using Keras.js. The accompanying GitHub repo contains the data and notebook needed to train. If I were to re-do this project, I would write the training procedure into a library, not a notebook.

Moaney

A minimal viable text classification app built around a fictional business problem. I use this for teaching and demonstrating how I architect small machine learning projects, following a Functional Core, Imperative Shell pattern: creating a library of reusable, functional components, and keeping stateful operations to short scripts.

hey guys

The tiniest Flask app, because I wanted to play with deploying to Heroku.

Unit Testing with PySpark

A very tiny project documenting the result of the longer-than-it-should-have-taken to set up a working PySpark testing environment, creating a new Spark context only once per full test run.