FastForest multithreading

[sretz]

System Information (please complete the following information):

• OS & Version: Windows 10 19044.1706
• ML.NET Version: 1.7.1 x64
• .NET Version: v4.8 x64

Describe the bug I'm running tests with FastForest on machine with 2 processors with 12 cores each with a total of 48 threads, the program running the test is using 48 threads, but the highest CPU the program gets is 14%, which 7 maybe 8 threads. This leads me to believe the FastForest training algo is doing some locking.

To Reproduce Steps to reproduce the behavior:

1. Create as many threads as the machine has
2. var trainer = context.BinaryClassification.Trainers.FastForest(label, numberOfTrees: trees);
3. var trained = trainer.Fit(inputs);
4. look at CPU usage

Expected behavior CPU usage at 100% not significantly less

Screenshots, Code, Sample Projects

Additional context

[michaelgsharp]

how many trees are you using? Is there anything else you are doing in your code besides what you listed here?

[sretz]

I am using 2 through 21 trees. Here is what I'm doing on the thread.

[tarekgh]

It is possible to be the lock https://github.com/dotnet/machinelearning/blob/9652e59c5d639cd30390fcfd47b525eb0bd8665e/src/Microsoft.ML.FastTree/FastTree.cs#L217