peopledoc / mlvtools

Public repository for versioning machine learning data
Other
42 stars 7 forks source link

Reiterate runs for uncertainty intervals #50

Open SdgJlbl opened 5 years ago

SdgJlbl commented 5 years ago

Building up on feature https://github.com/peopledoc/ml-versioning-tools/issues/48, if the experiment contains some stochasticity (random train / validation split, random initalization, stochastic gradient descent), we want to reiterate several runs of the experiments and to save the resulting metrics. Theses metrics values can be exported as a raw csv file or in a aggregate way (to estimate uncertainty).