Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update to the latest ML.NET 0.11.0-preview-27428-4. #1

Merged
merged 3 commits into from
Mar 6, 2019

Conversation

eerhardt
Copy link
Contributor

@eerhardt eerhardt commented Mar 4, 2019

The biggest issue to overcome was in BinaryClassification. Two trainers (Perceptron and Linear SVM) don't produce Probabilities, which means they aren't "calibrated" and can't get a evaluated as such. With that, it means their metrics don't produce Entropy.

cc @XamlBrewer

The biggest issue to overcome was in BinaryClassification. Two trainers (Perceptron and Linear SVM) don't produce Probabilities, which means they aren't "calibrated" and can't get a evaluated as such. With that, it means their metrics don't produce Entropy.
@XamlBrewer
Copy link
Owner

Wow. This is impressive! Thank you very much!
Please give me some time to digest this...

@eerhardt Wat would be a good time to merge this PR? I'll probably wait for your next official NuGet packages (ML and DataView), right?

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 5, 2019

Yes, I think if you'd like to wait for v0.11, which will be coming this week. Once v0.11 has shipped officially on nuget.org, I can update this PR to that released version. Then you would be free to merge.

But please feel free to give PR feedback now. The v0.11 APIs won't be changing from the version I am using - it is the "RC" build of v0.11. So the only known change would be to update the version from the preview to 0.11.0.

@XamlBrewer
Copy link
Owner

Is the heat map OK in your project? I have a lot of NaN values in the input matrix (a lot in Fare and all of the Age values). Most of these values are correctly filled out in the data set.

It looks like the DataView does not react properly to missing values, or to the commas in the passenger name. But maybe I'm just running against an invalid combination of experimental NuGet packages (ML, DataView, CpuMath).

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 5, 2019

No, I don't think the heat map is OK:

image

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 5, 2019

Looks like the problem was that we switched AllowQuoting from defaulting to true to now default to false with dotnet/machinelearning@ec418e4.

I added AllowQuoting = true, and the map now looks like:

image

AllowQuoting defaults to `false` now on the latest ML.NET.
@XamlBrewer
Copy link
Owner

A build in Release mode fails with "ilc.exe exited with code 1200". Is that because your daily packages at https://dotnet.myget.org/feed/dotnet-core/package/nuget/Microsoft.ML are debug builds? Or am I'm missing something?

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 5, 2019

I'm not sure. I tried looking into it and saw that it was raising an OutOfMemory error during ilc.exe. The daily builds of ML.NET are Release builds.

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 5, 2019

@XamlBrewer - I verified this is the same behvaior as building in Release mode in the master branch. It doesn't appear to be caused by my changes.

Are you seeing it in the master branch?

@eerhardt
Copy link
Contributor Author

eerhardt commented Mar 6, 2019

@XamlBrewer - ML.NET v0.11.0 is now on nuget.org. I've pushed a commit to update to the released version.

@XamlBrewer
Copy link
Owner

When you create a new UWP project in Visual Studio and hook the new NuGet v0.11.0 ML.NET to it, you also get the ilc.exe fail on a Release build.

@XamlBrewer XamlBrewer merged commit 720e3c6 into XamlBrewer:master Mar 6, 2019
@eerhardt eerhardt deleted the UpdateToLatestMLNET branch March 6, 2019 15:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants