search

airoldilab / ai-sgd

Towards stability and optimality in stochastic gradient descent
5 stars 2 forks source link

Remove need for defining sgd/batch in exp_normal_n5p2.R #6

Open dustinvtran opened 9 years ago

dustinvtran commented 9 years ago

sgd.R should be generic enough to work for all experiments.

ptoulis commented 9 years ago

Yes i think we need to have one code sgd.R for the SGD implementations and one examplesmain.R having generic code for the experiments -- the exp* scripts should be a few lines long and should be there to let others run the experiments quickly.

dustinvtran commented 9 years ago

run.all() in exp_poisson_n4p1.R is close to achieving that goal and will eventually be extended and placed into another file. There should also be more diagnostic functions aside from plot.risk(), such as a wrapper for performing time benchmarks.

ptoulis commented 9 years ago

Agreed. Let' s keep working on the other code, and eventually we will shape up the exp scripts when the rest is finalized.

dustinvtran commented 9 years ago

e0b9cd2b8f8e641ff1584070230ed5aba5829c9a cleans this as much as possible.