def test_save_load(self):
model = nnRegressor.Convolutional_NN_Model(self.input_dim,
self.target_dim)
reg = nnRegressor.NeuralNetRegressor(model=model,
patience=1,
use_gpu=True)
reg.save('testCNN')
self.assertTrue(os.path.exists('testCNN'))
newReg = nnRegressor.NeuralNetRegressor()
newReg.load('testCNN')
self.assertEquals(newReg.model.input_dim, self.input_dim)
self.assertEquals(newReg.model.output_dim, self.target_dim)
def test_save_load(self):
model = nnRegressor.Linear_NN_Model(self.input_dim, self.target_dim,
'mean')
reg = nnRegressor.NeuralNetRegressor(model=model,
patience=1,
use_gpu=True)
reg.save('test')
self.assertTrue(os.path.exists('test'))
newReg = nnRegressor.NeuralNetRegressor()
newReg.load('test')
self.assertEquals(newReg.model.fc1.in_features, self.input_dim)
self.assertEquals(newReg.model.fc3.out_features, self.target_dim)
def test_score(self):
model = nnRegressor.Linear_NN_Model(self.input_dim, self.target_dim,
'mean')
reg = nnRegressor.NeuralNetRegressor(model=model,
patience=1,
use_gpu=True)
fitted = reg.fit(self.X_train, self.y_train)
score = fitted.score(self.X_test, self.y_test)
def test_score(self):
input_dim = len(self.X_train[0, :])
target_dim = len(self.y_train[0, :])
model = nnRegressor.Convolutional_NN_Model(input_dim, target_dim)
fittedReg = nnRegressor.NeuralNetRegressor(
model=model, patience=1).fit(self.X_train, self.y_train)
score = fittedReg.score(self.X_test, self.y_test)
def test_predict_without_GPU_default_model(self):
fittedReg = nnRegressor.NeuralNetRegressor(patience=1).fit(
self.X_train, self.y_train)
result = fittedReg.predict(self.X_test)
self.assertEqual(next(fittedReg.model.parameters()).is_cuda, False)
self.assertEqual(result.shape,
(len(self.X_test), len(self.y_test[0, :])))
self.assertTrue(type(result) is numpy.ndarray)
self.assertTrue(result.dtype is numpy.dtype('float32')
or result.dtype is numpy.dtype('float64'))
def test_predict_without_GPU(self):
model = nnRegressor.Linear_NN_Model(self.input_dim, self.target_dim,
'mean')
fittedReg = nnRegressor.NeuralNetRegressor(
model=model, patience=1).fit(self.X_train, self.y_train)
result = fittedReg.predict(self.X_test)
self.assertEqual(next(fittedReg.model.parameters()).is_cuda, False)
self.assertEqual(result.shape,
(len(self.X_test), len(self.y_test[0, :])))
self.assertTrue(type(result) is numpy.ndarray)
self.assertTrue(result.dtype is numpy.dtype('float32')
or result.dtype is numpy.dtype('float64'))
def perform_GridSearch(X, y):
scorer = make_scorer(metrics.average_relative_root_mean_squared_error,
greater_is_better=False)
parameters = {
"learning_rate": [0.01], # 0.001, 0.1, 0.2, 0.3],
"patience": [1] #, 3, 5, 8]
}
reg = GridSearchCV(nn.NeuralNetRegressor(training_limit=10000),
parameters,
cv=3,
scoring=scorer,
n_jobs=1)
return reg.fit(X, y)
def test_predict_with_GPU(self):
if torch.cuda.is_available():
input_dim = len(self.X_train[0, :])
target_dim = len(self.y_train[0, :])
model = nnRegressor.Convolutional_NN_Model(input_dim, target_dim)
fittedReg = nnRegressor.NeuralNetRegressor(model=model,
patience=1,
use_gpu=True).fit(
self.X_train,
self.y_train)
result = fittedReg.predict(self.X_test)
self.assertEqual(next(fittedReg.model.parameters()).is_cuda, True)
self.assertEqual(result.shape,
(len(self.X_test), len(self.y_test[0, :])))
self.assertTrue(type(result) is numpy.ndarray)
self.assertTrue(result.dtype is numpy.dtype('float32')
or result.dtype is numpy.dtype('float64'))