def fit_categorical_labels(df_train, df_test, df_labels,
fit_type='regressor', fit_framework='theanets', labels_list=None):
from rep.estimators import SklearnClassifier, SklearnRegressor
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.ensemble import GradientBoostingRegressor
from rep.estimators.neurolab import NeurolabRegressor
from rep.estimators.theanets import TheanetsRegressor
#from rep.estimators import XGBoostRegressor
#from rep.estimators import XGBoostRegressor
# Using gradient boosting with default settings
if fit_framework == 'sklearn':
if fit_type == 'classifier':
sk = SklearnClassifier(GradientBoostingClassifier(),
features=df_train.columns.values)
elif fit_type == 'regressor':
sk = SklearnRegressor(GradientBoostingRegressor(),
features=df_train.columns.values)
elif fit_framework == 'neural':
if fit_type == 'regressor':
sk = NeurolabRegressor(features=df_train.columns.values,
)
elif fit_framework == 'xgboost':
if fit_type == 'regressor':
sk = XGBoostRegressor(features=df_train.columns.values,
)
elif fit_framework == 'theanets':
if fit_type == 'regressor':
sk = TheanetsRegressor(features=df_train.columns.values,
)
else:
raise ValueError('No correct combo of fit_type and fit_framework found')
prediction_array = np.empty((len(df_test), len(df_labels.columns)))
for i, column in enumerate(df_labels.columns.values):
# get a single column to predict
labels = df_labels[column]
# fit the data with the training set
sk.fit(df_train, labels)
# predict new countries
prediction = np.squeeze(sk.predict(df_test))
prediction_array[:, i] = prediction
#prediction = pd.read_pickle(filename).squeeze()
df_predict = pd.DataFrame(prediction_array, columns=df_labels.columns.values)
df_predict = gather_dummy_predictions(df_predict, labels_list)
#print('unique labels', np.unique(df_predict))
return df_predict
