Repository navigation
We need to find way to control random seed and amount of threads at least for tests. #135
Description
Currently out pipeline.Train function creates
using (var environment = new TlcEnvironment())
which is nice, but it doesn't specify seed which become random and concurrency which force algorithms to use all threads available on machine. Which can result in different metrics, predictions and any other stuff we assert in tests.
Reacted by Gleb K and Tom Finley
Activity
Ivanidzo4ka commentedon May 11, 2018
https://ci2.dot.net/job/Private/job/dotnet_machinelearning/job/master/job/windows_nt_release_prtest/588/testReport/junit/(root)/(empty)/Microsoft_ML_Scenarios_ScenariosTests_TrainAndPredictIrisModelWithStringLabelTest/
as example.
GalOshri commentedon May 11, 2018
Related to #9
TomFinley commentedon May 11, 2018
Hmm, yes. In this case, I suppose,
RandandConcurrencyFactor.Hi @glebuk , hi @KrzysztofCwalina ... here's another issue to add to the pile of those that could be easily addressed just by exposing the existing functionality of
IHostEnvironment.shauheen commentedon Jun 28, 2018
closing as some solution was merged and the underlying issue will be handled differently after #371 gets implemented.