Rewrite regression to use tensorflow

better / convoys

Implementation of statistical models to analyze time lagged conversions

https://better.engineering/convoys/

MIT License

254 stars 42 forks source link

Rewrite regression to use tensorflow #10

Closed erikbern closed 6 years ago

erikbern commented 6 years ago

Turns out tensorflow has implementations of the regularized incomplete gamma functions (tf.igamma) and all kinds of goodies that are already built with automatic differentiation and numeric stability and all that jazz.

Using a SGD-style optimizer also makes it a lot easier to use minibatches and stuff like that for large datasets.

Was pretty surprised how easy this was... did it in less than 2h (this is the first time ever using tensorflow too)

erikbern commented 6 years ago

Plan forward

[X] merge this
[ ] use this for some experiments at work and make sure the gamma implementation works
[ ] figure out what to do with the kaplan-meier model
[ ] drop all old models in __init__.py and replace with regression models

coveralls commented 6 years ago

Coverage increased (+1.3%) to 70.233% when pulling 40c84aa0791d0596e4ddb630f4288dfc2057e2fb on regression-tensorflow into 2dbc6d8076fffa00308b221e7e1c6e0025c06a77 on master.