From 7d3c766db94846a33cf89bb99255f3d0346001de Mon Sep 17 00:00:00 2001 From: Hyrum Anderson Date: Thu, 11 Jul 2019 22:10:08 +0000 Subject: [PATCH] expanded grid --- ember/__init__.py | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-) diff --git a/ember/__init__.py b/ember/__init__.py index 668b42a1..9c84c153 100644 --- a/ember/__init__.py +++ b/ember/__init__.py @@ -171,19 +171,18 @@ def optimize_model(data_dir): # define search grid param_grid = { - 'learning_rate': [0.005, 0.05], - 'n_estimators': [100, 500], - 'num_leaves': [32, 128, 512], + 'learning_rate': [0.05, 0.1], + 'num_iterations': [1000], + 'num_leaves': [1024,2048], + 'min_data_in_leaf': [50,100], + 'max_depth': [11,15], 'boosting_type': ['gbdt'], 'objective': ['binary'], - 'colsample_bytree': [0.8, 1.0], - 'subsample': [0.8, 1.0], - 'reg_alpha': [1, 1.2], - 'reg_lambda': [1, 1.2], + 'colsample_bytree': [0.5, 0.75, 1.0], # aka feature_fraction } model = lgb.LGBMClassifier( - boosting_type='gbdt', + boosting='gbdt', n_jobs = -1, silent = True ) @@ -192,7 +191,7 @@ def optimize_model(data_dir): # so this works for progrssive time series splitting progressive_cv = TimeSeriesSplit( n_splits=3 ).split(X_train) - grid = GridSearchCV(estimator=model, cv=progressive_cv, param_grid=param_grid, scoring=score, n_jobs=1, verbose=3) + grid = GridSearchCV(estimator=model, cv=progressive_cv, param_grid=param_grid, scoring=score, n_jobs=1, verbose=1) grid.fit( X_train, y_train )