bug: When all class_ids in the constructed TreeEnsembleClassifier object are 0, the returned labels are incorrect and the order of scores for each row is inconsistent.

[Cc-a23187]

Bug Report

Orion version:
v0.1.8

Scarb version:
scarb 2.3.1 (0c8def3aa 2023-10-31)
cairo: 2.3.1 (https://crates.io/crates/cairo-lang-compiler/2.3.1)
sierra: 1.3.0

Current behavior:

When all class_ids in the TreeEnsembleClassifier object are 0 (there are two cases for classlabels), the predicted results from the Orion operator differ from the predictions of the correct model. The scores are the same but the order is inconsistent.

Orion execute result in cairo

Expected behavior:

Get the same result as onnxruntime, like here👇:

Onnxruntime execute result in python

Steps to reproduce:
bc.zip

upzip bc.zip
cd bc
scarb cairo-run --available-gas 99999999999

Other information:

Problem Localization:

The TreeEnsembleClassifier::predict function in the Orion library is missing the [binary value processing] operation.

Solution:

Adding the corresponding binary logic will solve this issue. I will submit a PR today to address this issue.

[Cc-a23187]

Close the issue, notice the mention in this line