Treelite model output shape will be know only after the model run.
If model uses objective multi:softmax then output shape will be (batch, 1) because predictor will execute predictor_transform max_index.
mnist model uses multi:softmax objective which outputs one class with the maximum probability.
Treelite model output shape will be know only after the model run. If model uses objective multi:softmax then output shape will be (batch, 1) because predictor will execute predictor_transform max_index.
mnist model uses multi:softmax objective which outputs one class with the maximum probability.
Related discussion: https://github.com/dmlc/treelite/issues/351