This is a code description of Kaggle Competiton "Jane Street"
The project CODE and ideas are original development, but also follow the Apache Open Source License. Feel free to share correctly.
To build this network stucture, moslty we have to import several toolkits or package. All of them are showm in the notebook file.
Overall code was written in Python using Google Colab Pro which has several paid GPUs and privilage CPU for training.
The Main_ResNet file contains two frameworks which are the sophiticated strcture built by PyTorch and a simpleNN using TensorFlow while both two models show strong predictive power for this training set.
The other two .ipynb files is written about embeddings and blending which also had great performance for the dataset.
You can find the introduction and data desciption of the competition through the URL below.
https://www.kaggle.com/c/jane-street-market-prediction
The dateset contains an anonymized set of features. Several files are used as train, test and submission for the competition.
- Files
- train.csv
- example_test.csv
- example_sample_submission.csv
- feature.csv
My best ranking in this tournament is 60th, if I stabilize above and below this ranking in the end I will be well on my way to a SILVER medal. But unfortunately there was a "Black Swan" incidents halfway through the tournament, or maybe there was a problem with the network structure I built, causing my ranking to drop to as low as 2,000 places and beyond. I will upload part of my working code for backup, maybe I can find out the reason of my failure from here later.
Any extra information for these codes can contact me by email: [email protected]