Accelerated Multi-output Xgboost shows wrong results.

[lk1983823]

I am training a xgboost model with multiple inputs and outputs. The output feature_dim of the model is 11. Then I use d4p to accelerate the trained model to improve inference performance. Here is my code below:

daal_model = d4p.get_gbt_model_from_xgboost(xgbmodel.get_booster())
daal_prediction = d4p.gbt_regression_prediction().compute(test_x, daal_model)
logits = daal_prediction.prediction

But when I print the logits, I find the its output shape is (272303, 1). The outputs has only 1 feature_num, and the result is also wrong.
Thanks for the help.

Environment:

• OS: [e.g. Ubuntu 18.04]
• Compiler: [e.g. GCC9.2]
• Version: [e.g. 2019 Update 4]

[napetrov]

@lk1983823 thanks for report.

In short - multioutput is not supported

Will hold this issue opened for tracking documentation update and model builders update to warn on not supported multioutput models

[napetrov]

#1445 - documentation update

[david-cortes-intel]

Under the new interface for serving GBT models, this now throws an explicit error instead of producing incorrect results.