def objective03(automator, space):
'''
Objective function for Random Forest Regressor.
'''
algo = 'RandomForestRegressor'
X = automator.x_train
Y = automator.y_train
#Define the subset of dictionary keys that should get passed to the machine learning
#algorithm.
keys = get_keys(algo)
subspace = {k: space[k] for k in set(space).intersection(keys)}
#Extract the remaining keys that are pertinent to data preprocessing.
model = RandomForestRegressor(**subspace)
scaler = space.get('scaler')
num_features = space.get('k_best')
#Assemble a data pipeline with the extracted data preprocessing keys.
pipeline = []
pipeline = Pipeline([
('scaler', scaler),
('select_best', SelectKBest(k=num_features)),
('classifier', model),
])
#perform two passes of 10-fold cross validation and return the mean score.
kfold = RepeatedKFold(n_splits=10, n_repeats=1)
scores = -cross_val_score(pipeline,
X,
Y,
cv=kfold,
scoring=automator.score_metric,
verbose=False).mean()
return scores, algo
def objective04(automator, space):
'''
Objective function for Support Vector Machines. Note that this method uses a Bagged Classifier
as a wrapper for SVC. Support Vector Machine run time scales by O(N^3). Using bagged classifiers
break up the dataset into smaller samples so that runtime is manageable.
'''
algo = 'SVR'
X = automator.x_train
Y = automator.y_train
#Define the subset of dictionary keys that should get passed to the machine learning
#algorithm.
keys = get_keys(algo)
subspace = {k: space[k] for k in set(space).intersection(keys)}
#Build a model with the parameters from our Hyperopt search space.
n_estimators = space.get('n_estimators')
model = BaggingRegressor(
SVR(**subspace),
max_samples=automator.num_samples // n_estimators,
n_estimators=n_estimators,
)
scaler = space.get('scaler')
num_features = space.get('k_best')
#Assemble a data pipeline with the extracted data preprocessing keys.
pipeline = []
pipeline = Pipeline([
('scaler', scaler),
('select_best', SelectKBest(k=num_features)),
('classifier', model),
])
#perform cross validation and return the mean score.
kfold = RepeatedKFold(n_splits=automator.num_cv_folds,
n_repeats=automator.repeats)
scores = -cross_val_score(pipeline,
X,
Y,
cv=kfold,
scoring=automator.score_metric,
verbose=False,
n_jobs=-1).mean()
return scores, algo
def get_model(algo, space_dict):
keys = get_keys(algo)
space = {k: space_dict[k] for k in set(space_dict).intersection(keys)}
model_lib = {
'xgboost_classifier': XGBClassifier,
'xgboost_regressor': XGBRegressor,
'SGDClassifier': SGDClassifier,
'SGDRegressor': SGDRegressor,
'RandomForestClassifier': RandomForestClassifier,
'RandomForestRegressor': RandomForestRegressor,
'SVC': SVC,
'SVR': SVR,
'LogisticRegression': LogisticRegression,
'KNeighborClassifier': KNeighborsClassifier,
'KNeighborRegressor': KNeighborsRegressor,
'GaussianNB': GaussianNB,
}
return model_lib[algo](**space)
def objective01(automator, space):
'''
Objective function for XGBoost Classifier.
'''
algo = 'xgboost_classifier'
X = automator.x_train
Y = automator.y_train
#Define the subset of dictionary keys that should get passed to the machine learning
#algorithm.
keys = get_keys(algo)
subspace = {k:space[k] for k in set(space).intersection(keys)}
#Extract the remaining keys that are pertinent to data preprocessing.
model = XGBClassifier(n_jobs=-1, **subspace)
scaler = space.get('scaler')
num_features = space.get('k_best')
#Assemble a data pipeline with the extracted data preprocessing keys.
pipeline = []
pipeline = Pipeline([
('scaler', scaler),
('select_best', SelectKBest(k = num_features)),
('classifier', model),
])
#perform cross validation and return the mean score.
kfold = RepeatedKFold(n_splits = automator.num_cv_folds, n_repeats = automator.repeats)
try:
scores = -cross_val_score(pipeline, X, Y, cv=kfold, scoring = automator.score_metric, verbose=False).mean()
except ValueError:
print('An error occurred with the following space: ')
print(space)
return automator.best, algo
return scores, algo
def objective05(automator, space):
'''
Objective function for K-Nearest Neighbors Voting Regressor.
'''
algo = 'KNeighborRegressor'
X = automator.x_train
Y = automator.y_train
#Define the subset of dictionary keys that should get passed to the machine learning
#algorithm.
keys = get_keys(algo)
subspace = {k: space[k] for k in set(space).intersection(keys)}
#Build a model with the parameters from our Hyperopt search space.
model = KNeighborsRegressor(n_jobs=-1, **subspace)
scaler = space.get('scaler')
num_features = space.get('k_best')
#Assemble a data pipeline with the extracted data preprocessing keys.
pipeline = []
pipeline = Pipeline([
('scaler', scaler),
('select_best', SelectKBest(k=num_features)),
('classifier', model),
])
#perform cross validation and return the mean score.
kfold = RepeatedKFold(n_splits=automator.num_cv_folds,
n_repeats=automator.repeats)
scores = -cross_val_score(pipeline,
X,
Y,
cv=kfold,
scoring=automator.score_metric,
verbose=False).mean()
return scores, algo
def test_get_keys(self):
for key in ALGORITHM_KEYS.keys():
self.assertIsNotNone(get_keys(key))
print(get_keys(key))
