search

better / convoys

Implementation of statistical models to analyze time lagged conversions
https://better.engineering/convoys/
MIT License
258 stars 42 forks source link

Get rid of TensorFlow + use proper MCMC for uncertainty #55

Closed erikbern closed 6 years ago

erikbern commented 6 years ago

Don't think I got a lot of use out of TensorFlow. Once I started looking into doing proper MCMC I realized emcee does exactly what I want and it makes more sense to switch back to autograd+scipy.optimize to fit the curves

erikbern commented 6 years ago

this is incredibly slow in travis, not sure why, on my laptop this is as fast as tensorflow (if not faster)

coveralls commented 6 years ago

Coverage Status

Coverage decreased (-2.03%) to 89.899% when pulling acb26daf2cc767fdbb87ccd78d58a71d916f9d15 on nuke-tensorflow into d9697a27aeee0279ab2103049c8110af7b9b839e on master.