12th solution for the Otto Group Product Classification Challenge on Kaggle.
- NN3: Neural Networks with 3 hidden layers
- NN4: Neural Networks with 4 hidden layers. The first is sparse. The link is defined by RI or TRI.
- XGB: XGBoost
Feature extraction methods
- NewX = X * R, X:(N, 93), R:(93, m)
- Random Indexing(RI): RI controls # of nonzero elements in each row of R.
- Transposed Random Indexing(TRI): TRI controls # of nonzero elements in each column of R.
- Count Feature(CF): The count of each feature value
- TRI(k_max=4) + NN3
- TRI(k_max=5) + NN3
- NN4(RI)
- XGB + CF
- XGB + RI
- Ubuntu 14.04 LTS
- Cuda 6.5
- python 2.7
- numpy 1.8.2
- scipy 0.13.3
- scikit-learn 0.16.1
- theano 0.7
- pylearn2 0.1dev
- xgboost-0.3
- Put all *.py files into the folder containing train.csv, test.csv, sampleSubmission.csv
- python otto_nn.py
- Output files: pred_TRI_kmax_4.npy, pred_TRI_kmax_5.npy, pred_Sparse_RI.npy
- python otto_xgb.py
- Output files: pred_XGB_CF.npy, pred_XGB_RI.npy
- python make_submission.py
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent