Reguralized Selective ensemble

データの用意

train_test_split → data/data_n/ 以下に保存 n:0-4
- 最終的な評価のための5fold
- Input: all.data
- Output: data_n/X_test.npy X_train.npy y_test.npy y_train.npy
1 のtrainデータを5分割
- regularized selective ensemble の lambda 決定のための5fold
- train_test_split → data/data_n/fold_n 以下に保存 n:0-4
- Input: data_n/X_test.npy X_train.npy y_test.npy y_train.npy
- Output: data_n/fold_n/X_test.npy X_train.npy y_test.npy y_train.npy
アンサンブル候補モデルの予測を保存
- regularized selective ensemble の lambda 決定のための5fold
- Input: data_n/fold_n/X_test.npy X_train.npy y_test.npy y_train.npy
- Output: data_n/fold_n/predictions.csv truelabel.csv
アンサンブル候補モデルの予測を保存
- 3 と同じモデル群であること
- Input: data_n/X_test.npy X_train.npy y_test.npy y_train.npy
- Output: data_n/predictions.csv truelabel.csv

get_kernel_link_matrix.py 実行
- fold毎の w_link の計算
- data毎の w_link の計算
- Input: data_n/fold_n/X_test.npy data_n/X_test.npy
- Output: data_n/w_link/w_link_n.csv data_n/fold_n/w_link/w_link_n.csv
nohup echo " weight_for_train_data()" | matlab -nodisplay > weight.out & 実行
- train に対する weight の計算
- Input: data_n/fold_n/predictions.csv w_link/w_link_n.csv
- Output: data_n/fold_n/weight/weight_lambda_n.csv
lambda_decision.py 実行
- data毎の lambda の決定
- Input: data_n/fold_n/predictions.csv weight/weight_lambda_n.csv
- Output: lambda for each data
eval "matlab -nodesktop -nosplash -r \"weight_for_test_data($lambda)\" > weight.out" 実行
- train で決定した lambda に対して test に対する weight の計算
- Input: data_n/predictions.csv w_link/w_link_n.csv
- Output: data_n/weight/weight_lambda_n.csv

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
calc_weight		calc_weight
data		data
w_link		w_link
.gitignore		.gitignore
README.md		README.md
check_result.ipynb		check_result.ipynb
result.py		result.py
rse_pseudo-code.png		rse_pseudo-code.png