Info

Project was developed on mac os, tested on linux x86 Used python 3.9.1

Contest: https://contest.com/docs/ML-Competition-2023

Contest first results: https://t.me/contest/341

This project line:

    | Entry | Result | Time  
----+-------+--------+-------
6.  | 4672  | 74.5%  | 265ms

Build

cd src
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
cmake --build .

Train

1. Run dataset-prepare.ipynb to prepare dataset from raw_data_v2.zip. In result: 50000+ batches in dataset_v4o_train and 5000 batches in dataset_v4o_validate
2. Run dataset_tg-prepare.ipynb to prepare dataset from ml2023-r1-dataset.tar and ml2023-d1-dataset.tar
3. Run train.ipynb to train on data dataset_v4o batches. And save onnx model
4. Watch at convert-model.ipynb to convert onnx model to ort and compile onnxruntime