This repository contains codes for the Shopee-IET Machine Learning Kaggle Competition (Achieved rank 2 out of 51). The work was done during my internship at the Nanyang Technological University, Singapore in Spring 2018.
Contest page link: https://www.kaggle.com/c/shopee-iet-machine-learning-competition#description (Competition page have been removed.)
Leaderboard link (team name - hesl): https://www.kaggle.com/c/shopee-iet-machine-learning-competition/leaderboard (Competition page have been removed.)
Used a deep learning based ensemble model and a few data augmentation techniques to obtain top performance in the above challenge. Approach is explained in the presentation.pptx file
Prerequisites:
- Python 3
- Pytorch 0.3
Main scripts:
- train.py - Trains a model on the given data
- train_top5.py - Writes the top 5 predictions with their confidence for trained model on the training data.
- test.py - tests the model and outputs predictions
- test_top5.py - Predicts the top 5 classes along with their confidence on test data.
- ensemble_optimal.py - Uses the top 5 predictions of each model to give final prediction
Competition Snap:
Contributors:
- Punit Bhatt (Microsoft, India)
- Rydel Dcosta (Morgan Stanley, India)
- Omkar Damle (New York University, United States)