def tune_rbfsvr(): name = 'RbfSVR' dimension = 'length' stage, rbfsvr_reg, scale, features, rbfsvr_checkpoint_score = stage_meta_init(meta_dimension, name, dimension) if stage == 0: rbfsvr_reg = SVR(kernel = 'rbf') rbfsvr_checkpoint_score = -np.inf # features = init_feat_selection(X, Y, rbfsvr_reg) scale, rbfsvr_checkpoint_score = test_scaler(rbfsvr_reg, X, Y) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, list(X), final = False) elif stage == 1: rbfsvr_checkpoint_score, features = feat_selection_2(X[features], Y, scale, rbfsvr_reg, rbfsvr_checkpoint_score, 24, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, features, final = False) elif stage == 2: scale, linsvc_checkpoint_score = test_scaler(rbfsvr_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, features, final = False) elif stage == 3: rbfsvr_reg, rbfsvr_checkpoint_score = svc_hyper_parameter_tuning(X[features], Y, rbfsvr_reg, scale, rbfsvr_checkpoint_score) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, features, final = False) elif stage == 4: scale, rbfsvr_checkpoint_score = test_scaler(rbfsvr_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, features, final = False) elif stage == 5: rbfsvr_checkpoint_score, features = feat_selection_2(X[features], Y, scale, rbfsvr_reg, rbfsvr_checkpoint_score, 10, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, rbfsvr_reg, scale, rbfsvr_checkpoint_score, features, final = True)
def tune_polysvr(): name = 'PolySVR' dimension = 'length' stage, polysvr_reg, scale, features, polysvr_checkpoint_score = stage_meta_init(meta_dimension, name, dimension) if stage == 0: polysvr_reg = SVR(kernel = 'poly') polysvr_checkpoint_score = -np.inf # features = init_feat_selection(X, Y, rbfsvc_clf) scale, polysvr_checkpoint_score = test_scaler(polysvr_reg, X, Y) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, list(X), final = False) elif stage == 1: polysvr_checkpoint_score, features = feat_selection_2(X[features], Y, scale, polysvr_reg, polysvr_checkpoint_score, 24, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, features, final = False) elif stage == 2: scale, polysvr_checkpoint_score = test_scaler(polysvr_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, features, final = False) elif stage == 3: polysvr_reg, polysvr_checkpoint_score = poly_hyper_parameter_tuning(X[features], Y, polysvr_reg, scale, polysvr_checkpoint_score, iter_ = 50) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, features, final = False) elif stage == 4: scale, polysvr_checkpoint_score = test_scaler(polysvr_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, features, final = False) elif stage == 5: polysvr_checkpoint_score, features = feat_selection_2(X[features], Y, scale, polysvr_reg, polysvr_checkpoint_score, 10, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, polysvr_reg, scale, polysvr_checkpoint_score, features, final = True)
def tune_rf(): name = 'RFreg' dimension = 'length' stage, rf_reg, scale, features, rf_checkpoint_score = stage_meta_init(meta_dimension, name, dimension) if stage == 0: rf_reg = RandomForestRegressor(random_state = 1108, n_estimators = 100) rf_checkpoint_score = -np.inf scale, rf_checkpoint_score = test_scaler(rf_reg, X, Y) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, list(X), final = False) elif stage == 1: rf_checkpoint_score, features = feat_selection_2(X[features], Y, scale, rf_reg, rf_checkpoint_score, 24, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = False) elif stage == 2: scale, rf_checkpoint_score = test_scaler(rf_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = False) elif stage == 3: rf_reg, rf_checkpoint_score = forest_params(X[features], Y, rf_reg, scale, rf_checkpoint_score, iter_ = 1000) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = False) elif stage == 4: scale, rf_checkpoint_score = test_scaler(rf_reg, X[features], Y) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = False) elif stage == 5: rf_checkpoint_score, features = feat_selection(X[features], Y, scale, rf_reg, rf_checkpoint_score, 24, -1, False) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = False) elif stage == 6: rf_reg, rf_checkpoint_score = rf_trees(X, Y, scale, rf_reg, rf_checkpoint_score) _save_meta_model(meta_dimension, stage, dimension, name, rf_reg, scale, rf_checkpoint_score, features, final = True)
def tune_dartr(): name = 'DartrGBM' dimension = 'length' stage, dart_reg, scale, features, dartr_checkpoint_score = stage_init( name, dimension) if stage == 0: # import time # start = time.time() dart_reg = lgb.LGBMRegressor(random_state=1108, n_estimators=100, subsample=.8, verbose=-1) dartr_checkpoint_score = -np.inf scale, dartr_checkpoint_score = test_scaler(dart_reg, X, Y) # done = time.time() # done - start #ubuntu #sinlge: 50.78195834159851 #multi: 64.57199811935425 _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, list(X), final=False) elif stage == 1: dartr_checkpoint_score, features = feat_selection_2( X[features], Y, scale, dart_reg, dartr_checkpoint_score) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 2: scale, dartr_checkpoint_score = test_scaler(dart_reg, X[features], Y) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 3: dart_reg, dartr_checkpoint_score = lgb_find_lr(dart_reg, X[features], Y, scale, dartr_checkpoint_score) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 4: scale, dartr_checkpoint_score = test_scaler(dart_reg, X[features], Y) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 5: dartr_checkpoint_score, features = feat_selection( X[features], Y, scale, dart_reg, dartr_checkpoint_score, _iter=25) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 6: dart_reg, dartr_checkpoint_score = lgb_tree_params( X[features], Y, dart_reg, scale, dartr_checkpoint_score) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 7: scale, dartr_checkpoint_score = test_scaler(dart_reg, X[features], Y) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 8: dartr_checkpoint_score, features = feat_selection( X[features], Y, scale, dart_reg, dartr_checkpoint_score) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=False) elif stage == 9: dart_reg, dartr_checkpoint_score = lgb_drop_lr(dart_reg, X[features], Y, scale, dartr_checkpoint_score) _save_model(stage, dimension, name, dart_reg, scale, dartr_checkpoint_score, features, final=True)
def tune_lgr(): name = 'LightGBR' dimension = 'length' stage, lgb_reg, scale, features, lgbr_checkpoint_score = stage_init( name, dimension) if stage == 0: lgb_reg = lgb.LGBMRegressor(random_state=1108, n_estimators=100, subsample=.8, verbose=-1) lgbr_checkpoint_score = -np.inf scale, lgbr_checkpoint_score = test_scaler(lgb_reg, X, Y) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, list(X), final=False) elif stage == 1: lgbr_checkpoint_score, features = feat_selection_2( X[features], Y, scale, lgb_reg, lgbr_checkpoint_score) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 2: scale, lgbr_checkpoint_score = test_scaler(lgb_reg, X[features], Y) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 3: lgb_reg, lgbr_checkpoint_score = lgb_find_lr(lgb_reg, X[features], Y, scale, lgbr_checkpoint_score) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 4: scale, lgbr_checkpoint_score = test_scaler(lgb_reg, X[features], Y) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 5: lgbr_checkpoint_score, features = feat_selection( X[features], Y, scale, lgb_reg, lgbr_checkpoint_score) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 6: lgb_reg, lgbr_checkpoint_score = lgb_tree_params( X[features], Y, lgb_reg, scale, lgbr_checkpoint_score) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 7: scale, lgbr_checkpoint_score = test_scaler(lgb_reg, X[features], Y) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 8: lgbr_checkpoint_score, features = feat_selection(X[features], Y, scale, lgb_reg, lgbr_checkpoint_score, _iter=25) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=False) elif stage == 9: lgb_reg, lgbr_checkpoint_score = lgb_drop_lr(lgb_reg, X[features], Y, scale, lgbr_checkpoint_score) _save_model(stage, dimension, name, lgb_reg, scale, lgbr_checkpoint_score, features, final=True)
def tune_polysvc(): name = 'PolySVC' dimension = 'winner' stage, polysvc_clf, polysvc_checkpoint_score = stage_init( name, dimension, extension=EXTENSION) if stage == 0: polysvc_clf = SVC(random_state=1108, class_weight='balanced', kernel='poly', probability=True) polysvc_clf, polysvc_checkpoint_score = pipe_init(X, Y, polysvc_clf) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 1: polysvc_clf, polysvc_checkpoint_score = test_scaler( polysvc_clf, polysvc_checkpoint_score, X, Y) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 2: polysvc_clf, polysvc_checkpoint_score = feat_selection_2( X, Y, polysvc_clf, polysvc_checkpoint_score) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 3: polysvc_clf, polysvc_checkpoint_score = test_scaler( polysvc_clf, polysvc_checkpoint_score, X, Y) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 4: polysvc_clf, polysvc_checkpoint_score = pca_tune( X, Y, polysvc_clf, polysvc_checkpoint_score) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 5: polysvc_clf, polysvc_checkpoint_score = feat_selection_2( X, Y, polysvc_clf, polysvc_checkpoint_score) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 6: polysvc_clf, polysvc_checkpoint_score = test_scaler( polysvc_clf, polysvc_checkpoint_score, X, Y) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 7: polysvc_clf, polysvc_checkpoint_score = svc_hyper_parameter_tuning( X, Y, polysvc_clf, polysvc_checkpoint_score) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 8: polysvc_clf, polysvc_checkpoint_score = pca_tune( X, Y, polysvc_clf, polysvc_checkpoint_score, iter_=10) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 9: polysvc_clf, polysvc_checkpoint_score = feat_selection_2( X, Y, polysvc_clf, polysvc_checkpoint_score) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=False, extension=EXTENSION) elif stage == 10: polysvc_clf, polysvc_checkpoint_score = test_scaler( polysvc_clf, polysvc_checkpoint_score, X, Y) _save_model(stage, 'winner', name, polysvc_clf, polysvc_checkpoint_score, final=True, extension=EXTENSION)
def tune_knn(): name = 'KNN' dimension = 'winner' stage, knn_clf, knn_checkpoint_score = stage_init(name, dimension, extension=EXTENSION) if stage == 0: knn_clf = KNeighborsClassifier() knn_clf, knn_checkpoint_score = pipe_init(X, Y, knn_clf) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 1: knn_clf, knn_checkpoint_score = test_scaler(knn_clf, knn_checkpoint_score, X, Y) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 2: knn_clf, knn_checkpoint_score = feat_selection_2( X, Y, knn_clf, knn_checkpoint_score) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 3: knn_clf, knn_checkpoint_score = test_scaler(knn_clf, knn_checkpoint_score, X, Y) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 4: knn_clf, knn_checkpoint_score = pca_tune(X, Y, knn_clf, knn_checkpoint_score) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 5: knn_clf, knn_checkpoint_score = knn_hyper_parameter_tuning( X, Y, knn_clf, knn_checkpoint_score) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 6: knn_clf, knn_checkpoint_score = feat_selection_2( X, Y, knn_clf, knn_checkpoint_score) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 7: knn_clf, knn_checkpoint_score = test_scaler(knn_clf, knn_checkpoint_score, X, Y) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=False, extension=EXTENSION) elif stage == 8: knn_clf, knn_checkpoint_score = pca_tune(X, Y, knn_clf, knn_checkpoint_score, iter_=10) _save_model(stage, 'winner', name, knn_clf, knn_checkpoint_score, final=True, extension=EXTENSION)