Skip to content

Calculate ReduceSum row by row in ONNX model from OneVsAllTrainer #4904

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.

Sign up for GitHub

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

Merged
merged 1 commit into from
Mar 2, 2020
Merged

Calculate ReduceSum row by row in ONNX model from OneVsAllTrainer #4904

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions src/Microsoft.ML.StandardTrainers/Standard/MulticlassClassification/OneVersusAllTrainer.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -928,6 +928,8 @@ public override bool SaveAsOnnx(OnnxContext ctx, string[] outputNames, string fe
var sumOutput = ctx.AddIntermediateVariable(NumberDataViewType.Single, "SumOutput");
var sumNode = ctx.CreateNode(opType, expOutput, sumOutput, ctx.GetNodeName(opType), "");
sumNode.AddAttribute("keepdims", 1);
long[] list = { 1 };
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about the graph itself when run with SoftMax. Does it still look broken?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, the sigmoid nodes (that are actually never needed for lightgbm multiclass with softmax) are still there, without having their output connected to anything.

I did try to find how to eliminate them, but it seems that various changes are needed in onevsall trainer and also in the calibrators' saveasonnx methods.

I didn't look much into it, since i didn't know if it was worth it, as solving that won't have any effect on the model's output, and it actually isn't related to the issue i am addresing in this PR (if anything, it makes the onnx model architecture uglier and somewhat confusing, but nothing else).

Do you want me to fix that problem? In this PR?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, please fix the problem. I will leave it to you whether you want to fix this in a separate PR or this PR.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, I will address that in a separate PR.

sumNode.AddAttribute("axes", list);

opType = "Div";
var divOutput = ctx.AddIntermediateVariable(type, "DivOutput");
Expand Down