def get_trained_model(dataset,
label,
model_name,
task,
method_name='standard',
max_evals=100,
test_size=0.3,
random_state=1):
'''
Train the model with given data and hpo method . Returns the trained model
Parameters:
dataset(dataframe) : data to be used for training model
label (string): target column of the dataframe
task (string) : type of task
model_name(string) : name of the model on which data is to be trained
method_name(string) : Name of the hyper parameter method to be used while training
max_evals(int) : Number of evaluators
test_size(float) : Fraction of the data to be used for testing
random_state(int) : Random state to be used
Returns:
model (model object) : the trained model on which hpo is performed
'''
features = get_features(dataset, label)
X, Y = dataset[features], dataset[label]
X_train, X_test, Y_train, Y_test = train_test_split(
X, Y, test_size=test_size, random_state=random_state)
model = get_model(model_name)
if method_name == 'standard':
model = model()
trained_model = model.fit(X_train, Y_train)
print('Standard Model without HPO')
print('Model Trained')
return trained_model
elif method_name == 'grid_search':
trained_model = grid_search(model, X_train, Y_train)
elif method_name == 'random_search':
trained_model = random_search(model, X_train, Y_train)
# elif method_name == 'bayesian_gp':
# trained_model = bayesian_gp(model, X_train, Y_train)
elif method_name == 'bayesian_tpe':
trained_model = bayesian_tpe(model,
X_train,
X_test,
Y_train,
Y_test,
task,
max_evals=max_evals)
else:
raise Exception("'No hpo method named {}'.format(method_name)")
print('Hyperparameters Optimized')
print('Model Trained')
return trained_model
#---------------------------------------------------------------------------------------------------------------------#
def add_models(self, base_layer_models):
'''
Initializes the base layer models of the object
Parameters:
self (object ref)
base_layer_model (list) : list of models to be used at base layer in ensembling
'''
self.models.extend([get_model(model) for model in base_layer_models])
def __init__(self,
base_layer_models=None,
meta_model='RandomForestClassifier',
n_splits=5,
optimize=True,
max_evals=100):
'''
Initializes the object values with given arguments.
Parameters:
self (object ref)
base_layer_model (list) : list of models to be used at base layer in ensembling
meta_layer_model (string) : name of model to be used at meta layer in ensembling
n_splits(int) : number of splits to be made for cross validation
optimize(boolean) : optimize the process
max_evals(int) : max number of evaluations to be done
'''
if base_layer_models == None:
base_layer_models = [
'LogisticRegression', 'DecisionTreeClassifier',
'RandomForestClassifier', 'GradientBoostingClassifier',
'ExtraTreesClassifier', 'AdaBoostClassifier'
]
#base_layer_models = ['LogisticRegression', 'DecisionTreeClassifier']
self.models = [get_model(model) for model in base_layer_models]
if meta_model == None:
self.meta_model = None
elif meta_model in ['AdaBoostClassifier', 'BaggingClassifier']:
self.meta_model = get_model(meta_model)(
base_estimator=get_model('RandomForestClassifier')())
else:
self.meta_model = get_model(meta_model)()
self.n_splits = n_splits
self.trained_models = []
self.optimize = optimize
self.max_evals = max_evals