docker compose build
docker compose -f compose.cpu.yaml build
# bash に入る場合
docker compose run --rm kaggle bash # GPU
docker compose -f compose.cpu.yaml run --rm kaggle-cpu bash # CPU
# jupyter lab を起動する場合
docker compose up # GPU
docker compose -f compose.cpu.yaml up # CPUpython experiments/check/run.py exp=001
python experiments/check/run.py exp=base- 各スクリプトに共通する基本的な設定は yamls/config.yaml 内にある
- 各スクリプトによって変わる設定は、実行スクリプトのあるフォルダ(
{major_exp_name})の中にexp/{minor_exp_name}.yamlとして配置することで管理。- 実行時に
exp={minor_exp_name}で上書きする {major_exp_name}と{minor_exp_name}の組み合わせで実験が再現できるようにする
- 実行時に
- 準備: input/atmaCup16_Dataset にデータを置く
- 環境：GCE n2-standard-48 (48 vCPU, 24 core, 192 GB memory)
# 実行環境
docker compose -f compose.cpu.yaml run --rm kaggle-cpu bash
# 候補・特徴量作成
python cand_unsupervised/transition_prob_fix/run.py exp=base
python cand_unsupervised/transition_prob_all_fix/run.py exp=base
python cand_unsupervised/ranking_location/run.py exp=sml_cd
python cand_unsupervised/ranking_location/run.py exp=lrg_cd
python cand_unsupervised/ranking_location_all/run.py exp=sml_cd
python cand_unsupervised/split_transition_prob_fix/run.py exp=base
python cand_unsupervised/split_transition_prob_all_fix/run.py exp=base
python cand_unsupervised/split_transition_prob_bidirect_all_fix/run.py exp=base
python cand_unsupervised/split_feat_transition_prob_location/run.py exp=base
python cand_unsupervised/split_ranking/run.py exp=base
python cand_unsupervised/split_ranking_location/run.py exp=ken_cd
python cand_unsupervised/split_ranking_location/run.py exp=lrg_cd
python cand_unsupervised/split_ranking_location/run.py exp=sml_cd
python cand_unsupervised/split_ranking_location/run.py exp=wid_cd
python cand_unsupervised/prob_matrix_filter/run.py exp=two002
# fold 作成
python generate_datasets/make_cv/run.py
# 学習用データ生成
python generate_datasets/030_train_test_feat/run.py exp=base
python generate_datasets/030_train_test_feat/run.py exp=other
python generate_datasets/030_train_test_feat/run.py exp=first
# 学習：長さ1
python experiments/008_split/run.py exp=v030_first
# 学習：長さ1以外
python experiments/008_split/run.py exp=v030_other
python experiments/008_split/run.py exp=v030_other001
python experiments/012_cat_boost/run.py exp=v030_other_001
# 遷移行列の計算
python cand_supervised/prob_matrix_test_weight/run.py exp=021
# それぞれの予測値を結合してsubmissionファイルを作る
python experiments/ensemble_007/run.py exp=032