def run_over_all_input_parameters_reg(X, y, max_evals, cvs, val_sizes, methods, pos_labels, test_sizes, top_cv_evals, thresholds, eval_metrics):
random_state = 42
out = []
count = 0
for max_eval in max_evals:
for cv in cvs:
for val_size in val_sizes:
for method in methods:
for pos_label in pos_labels:
for test_size in test_sizes:
for top_cv_eval in top_cv_evals:
for threshold in thresholds:
for eval_metric in eval_metrics:
try:
status = 'OK'
loss = np.nan
# Setup model
hgb = hgboost(max_eval=max_eval, threshold=threshold, cv=cv, test_size=test_size, val_size=val_size, top_cv_evals=top_cv_eval, random_state=random_state, verbose=2)
# Fit model
if method=='xgb_reg':
hgb.xgboost_reg(X, y, eval_metric=eval_metric);
elif method=='ctb_reg':
hgb.catboost_reg(X, y, eval_metric=eval_metric);
elif method=='lgb_reg':
hgb.lightboost_reg(X, y, eval_metric=eval_metric);
# use the predictor
y_pred, y_proba = hgb.predict(X)
# Loss score
loss = hgb.results['summary']['loss'].iloc[np.where(hgb.results['summary']['best'])[0]].values
# Make some plots
# assert gs.plot_params(return_ax=True)
# assert gs.plot(return_ax=True)
# assert gs.treeplot(return_ax=True)
# if (val_size is not None):
# ax = gs.plot_validation(return_ax=True)
# assert len(ax)>=2
except ValueError as err:
assert not 'hgboost' in err.args
status = err.args
print(err.args)
tmpout = {'max_eval':max_eval,
'threshold':threshold,
'cv':cv,
'test_size':test_size,
'val_size':val_size,
'top_cv_evals':top_cv_eval,
'random_state':random_state,
'pos_label':pos_label,
'method':method,
'eval_metric':eval_metric,
'loss':loss,
'status':status,
}
out.append(tmpout)
count=count+1
print('Fin! Total number of models evaluated with different paramters: %.0d' %(count))
return(pd.DataFrame(out))
def get_data():
from hgboost import hgboost
gs = hgboost()
df = gs.import_example()
y = df['Parch'].values
y[y>=3]=3
del df['Parch']
X = gs.preprocessing(df, verbose=0)
return X, y
def get_data_reg():
from hgboost import hgboost
gs = hgboost()
df = gs.import_example()
y = df['Age'].values
del df['Age']
I = ~np.isnan(y)
X = gs.preprocessing(df, verbose=0)
X = X.loc[I,:]
y = y[I]
return X, y
# hyperparameter searching
# feature importance
# early stopping
# plotting
# %%
from hgboost import hgboost
print(dir(hgboost))
# print(hgboost.__version__)
import numpy as np
# %% HYPEROPTIMIZED XGBOOST
hgb_xgb = hgboost(max_eval=10,
threshold=0.5,
cv=5,
test_size=0.2,
val_size=0.2,
top_cv_evals=10,
random_state=None,
verbose=3)
hgb_cat = hgboost(max_eval=10,
threshold=0.5,
cv=5,
test_size=0.2,
val_size=0.2,
top_cv_evals=10,
random_state=None,
verbose=3)
hgb_light = hgboost(max_eval=10,
threshold=0.5,
cv=5,
test_size=0.2,