# "n_estimators": [1000, 1500],
# "max_depth": [5, 10],
# "max_features": [0.3, "sqrt"]
# }
cv_i = False
param_grid = {
"loss": "huber",
"learning_rate": 0.01,
"n_estimators": 1000,
"max_depth": 5,
"max_features": 10,
"warm_start": True,
}
gbm_CV, pred_test = gbm_fun.train_and_test_gradboost(training_features,
traj_testing,
param_grid=param_grid,
cv=cv_i)
np.save(saving_path + "predicted_test_set.npy", pred_test)
joblib.dump(gbm_CV, saving_path + "clf.pkl")
# predictions
if cv_i is True:
alg = gbm_CV.best_estimator_
ml.write_to_file_cv_results(saving_path + "cv_results.txt", gbm_CV)
else:
alg = gbm_CV
ada_r2_train = np.zeros(len(alg.estimators_), )
for i, y_pred in enumerate(alg.staged_predict(training_features[:, :-1])):
ada_r2_train[i] = r2_score(log_halo_training, y_pred)
# param_grid = {"loss": ["huber"],
# "learning_rate": [0.2],
# "n_estimators": [500, 800],
# "max_depth":[5, 100],
# "max_features":[0.3, "sqrt"]
# }
param_grid = {"loss": ["huber"],
"learning_rate": [0.1, 0.15, 0.2],
"n_estimators": [800, 1000],
"max_depth":[5],
"max_features":[0.3, "sqrt"]
}
ada_CV, pred_test = gbm_fun.train_and_test_gradboost(tr_features_07_04, testing_signal_07_corr,
param_grid=param_grid, cv=cv_i)
np.save(saving_path + "predicted_test_set.npy", pred_test)
ml.write_to_file_cv_results(saving_path + "cv_results.txt", ada_CV)
joblib.dump(ada_CV, saving_path + "clf.pkl")
# predictions
if cv_i is True:
alg = ada_CV.best_estimator_
else:
alg = ada_CV
ada_r2_train = np.zeros(len(alg.estimators_), )
for i, y_pred in enumerate(alg.staged_predict(tr_features_07_04[:, :-1])):
ada_r2_train[i] = r2_score(log_mass_training, y_pred)
"loss": ["huber"],
"learning_rate": [0.1],
"n_estimators": [800],
"max_depth": [5],
"max_features": [0.3, 0.2]
}
param_grid = {
"loss": "huber",
"learning_rate": 1,
"n_estimators": 800,
"max_depth": 5,
"max_features": 0.2
}
gbm, pred_test = gbm_fun.train_and_test_gradboost(training_features,
testing_features,
param_grid=param_grid,
cv=False)
np.save(saving_path + "predicted_test_set.npy", pred_test)
joblib.dump(gbm, saving_path + "clf.pkl")
np.save(saving_path + "importances.pdf",
gbm.best_estimator_.feature_importances_)
ml.write_to_file_cv_results(saving_path + "cv_results.txt", gbm)
# predictions
if cv_i is True:
alg = gbm.best_estimator_
else:
alg = gbm
z0_den_testing = z0_den_features[testing_set]
log_halo_testing = np.log10(halo_mass[testing_set])
np.save(saving_path_z0 + "log_halo_testing_set.npy", log_halo_testing)
# training z=0
param_grid = {
"loss": "lad",
"learning_rate": 0.06,
"n_estimators": 2000,
"max_depth": 5,
"max_features": "sqrt"
}
clf_z0, pred_test = gbm_fun.train_and_test_gradboost(training_features,
z0_den_testing,
param_grid=param_grid,
cv=False)
np.save(saving_path_z0 + "predicted_test_set.npy", pred_test)
joblib.dump(clf_z0, saving_path_z0 + "clf.pkl")
################# GBT VS RF #####################
pred_z0 = np.load(
"/Users/lls/Documents/mlhalos_files/regression/gradboost/random_sampled_training/z0_den"
"/nest_2000_lr006/predicted_test_set.npy")
true_z0 = np.load(
"/Users/lls/Documents/mlhalos_files/regression/gradboost/random_sampled_training/z0_den"
"/nest_2000_lr006/log_halo_testing_set.npy")
pred_RF_z0 = np.load(
"/Users/lls/Documents/mlhalos_files/regression/lowz_density/z0/z0_only/predicted_log_halo_mass.npy"