You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In ML.NET codebase, one can see two different names for 32-bit float type. Recent PR uses float while old places have Float. It'd be nice to have only one of the two styles.
The .NET Design guidelines indicate we should use the language keyword (float) everywhere except for API names, where we should use the non-language specific name (Single):
@wschin we are trying to get rid of Float in our code, and if you look into estimators PRs we almost always delete that using Float = System.Single; line and use float instead
Just some history, in the long vanished past we used to have two builds, a "double" and a "single" build. (This was in the days before IDataView where pipelines could accommodate either.) But, when we removed the compiler switch, it was beneficial to keep the declaration around: by doing so you only had a couple thousand lines changed, instead of tens of thousands of lines changed.
We've been opportunistically removing Float though, though in a very very piecemeal fashion. Doing it all at once though -- well, it might be a good idea, but that would be one heck of a PR to manage conflicts on. 😛
Activity
tannergooding commentedon Nov 19, 2018
The .NET Design guidelines indicate we should use the language keyword (
float) everywhere except for API names, where we should use the non-language specific name (Single):For example:
CC. @eerhardt
Ivanidzo4ka commentedon Nov 19, 2018
@wschin we are trying to get rid of
Floatin our code, and if you look into estimators PRs we almost always delete thatusing Float = System.Single;line and usefloatinsteadTomFinley commentedon Nov 19, 2018
Just some history, in the long vanished past we used to have two builds, a "double" and a "single" build. (This was in the days before
IDataViewwhere pipelines could accommodate either.) But, when we removed the compiler switch, it was beneficial to keep the declaration around: by doing so you only had a couple thousand lines changed, instead of tens of thousands of lines changed.We've been opportunistically removing
Floatthough, though in a very very piecemeal fashion. Doing it all at once though -- well, it might be a good idea, but that would be one heck of a PR to manage conflicts on. 😛Zruty0 commentedon Nov 20, 2018
Yes, just recently I saw someone mass-renaming
Singletofloat, and ending up withLdafloatboxsomewhere in comments :)glebuk commentedon Jan 14, 2019
Seems to be fixed already
Ivanidzo4ka commentedon Jan 14, 2019
Not exactly finished.
machinelearning/src/Microsoft.ML.StandardLearners/Optimizer/LineSearch.cs
Line 8 in a570da1
We have still plenty of places where we have
using Float = System.Single;pattern.In many places we clean it up via different unrelated PRs, but some code haven't been touch since we merge it from internal repo.
sfilipi commentedon Feb 6, 2019
The action item on this PR is to remove
using Float = System.Singlefrom all the files, and through the codebase, replaceFloat, withfloat.code migration from machinelearning-automl to machinelearning repo fe…
Random number generator is not thread safe (dotnet#310)
Added onnx export functionality for LpNormNormalizingTransformer (#4161)
Added Onnx export functionality to PCATransformer (#4188)