About

My developments for MTS ML CUP

Final solution: all features (aggregates (mean, median, min, max, std) by user_id (prices, counts, categorical features) + mean encoded (w2v + node2vec) url vectors) + 10 cv DANets averaging

Hardware:

CPU: Intel Xeon E5-2650V2

RAM: 64 Gb DDR3

GPU: RTX 3060 (12 Gb)

Installation

Place competition data into cameleogrey_mtsmlcup/data/raw

git clone https://github.com/CameleoGrey/cameleogrey_mtsmlcup.git
conda create -n mtsmlcup python=3.10
conda activate mtsmlcup
pip3 install torch==1.13.0+cu117 torchvision==0.14.0+cu117 torchaudio===0.13.0+cu117 -f https://download.pytorch.org/whl/cu117/torch_stable.html
pip install -r requirements.txt

If you're using Eclipse IDE with PyDev plugin, make this inside project properties: Properties --> Resource --> Text file encoding --> Other --> UTF-8

If you want to try scrapping / parsing features

pip install transformers sentence-transformers
pip install selenium

+ you need to install Chrome Driver for selenium

If you want to try graph features

pip install "tensorflow<2.11"

1. Download stellargraph package (https://github.com/stellargraph/stellargraph)
2. Place stellargraph into envs/mtsmlcup/Lib/site-packages
3. Move into stellargraph dir
4. Inside setup.py comment #python_requires=">=3.6.0, <3.9.0",
5. cd envs/mtsmlcup/Lib/site-packages/stellargraph
6. pip install .