AITM-torch

Support Python and C++

• Official Tensorflow implementation
• Paper

Dataset

Download and preprocess dataset use script in official Tensorflow implementation

python process_public_dataset.py

Train && Test

cd python
mkdir out
python train.py

AUC performance

Test AUC on Alibaba Click and Conversion Prediction dataset

Test Resutt: click AUC: 0.6189267022220789 conversion AUC:0.6544229866061039

Deployment

Pytorch c++ front is 2x faster than python version and 6x faster than TorchScript when inference. We Implement a C++ version that can load tochscript format file There is a demo cpp

build

Download libtorh libtorch installing docs

cd cpp
mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=/absolute/path/to/libtorch ..
cmake --build . --config Release