Skip to content

Latest commit

 

History

History
83 lines (50 loc) · 1.81 KB

README.md

File metadata and controls

83 lines (50 loc) · 1.81 KB
Raw

README

This project is under Apache 2.0 License.

App Inference Demo

prediction sample

Train

Directory: xgboost_train.

Without bazel

Just run train.ipynb.

Using bazel

  1. Install bazelisk.

You can install bazelisk in multiple ways, including:

  • npm install -g @bazel/bazelisk
  • using a binary release for Linux, macOS, or Windows
  • using Homebrew on macOS
  • by compiling from source using Go: `go get github.com/bazelbuild/bazelisk
  1. Run XGBoostClassifier with Iris dataset:
cd ./xgboost_train
bazelisk run :train

The output model is under xgboost_train/bazel-xgboost_train.

  1. Convert (.onnx to .ort) (optional)

If you need to use .ort model, you can convert with:

python -m onnxruntime.tools.convert_onnx_models_to_ort <onnx model file or dir>

But for xgboost model, .ort model is 2~3 times larger than original .onnx model.

Inference (on Android)

Directory: onnx_inference.

Install Android Studio and run the APK.

Q&A

Why do not deploy with XGBoost4j?

The Android architecture is not official supported in XGBoost4j.

Why do not convert onnx to TensorFlow and deploy with TFLite?

Because it's impossible as TensorFlow do not implement the XGBoost backend. If you convert onnx to tf model, you would get this convertion error:

Convert code:

import onnx
from onnx_tf.backend import prepare

# Load the ONNX model
onnx_model = onnx.load("xgbc_iris.onnx")

# Convert the ONNX model to TensorFlow format
tf_model = prepare(onnx_model)

# Save the TensorFlow model
tf_model.export_graph("xgbc_iris.tf")

Convert result (ERROR):

Error: "BackendIsNotSupposedToImplementIt: TreeEnsembleClassifier is not implemented."