def test12_optimizer_create_updates(self,mock_shared,mock_polyak,mock_adagrad,mock_constrain,mock_rmsprop,mock_sgd,mock_adam,mock_adadelta,mock_nesterov):
try:
mock_polyak.return_value = ""
mock_adagrad.return_value = ""
mock_constrain.return_value = ""
mock_rmsprop.return_value = ""
mock_sgd.return_value = ""
mock_adam.return_value = ""
mock_adadelta.return_value = ""
mock_nesterov.return_value = ""
mock_shared.return_value = self.learning_val
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt.optimizer_type ="adagrad"
self.opt.momentum_type ="polyak"
self.opt.create_updates(verbose=self.verbose)
self.opt.optimizer_type ="rmsprop"
self.opt.momentum_type ="nesterov"
self.opt.create_updates(verbose=self.verbose)
self.opt.optimizer_type ="sgd"
self.opt.momentum_type ="polyak"
self.opt.create_updates(verbose=self.verbose)
self.opt.optimizer_type ="adam"
self.opt.momentum_type ="polyak"
self.opt.create_updates(verbose=self.verbose)
self.opt.optimizer_type ="adadelta"
self.opt.momentum_type ="polyak"
self.opt.create_updates(verbose=self.verbose)
self.opt.optimizer_type = "else"
self.opt.create_updates(verbose=self.verbose)
self.assertEqual(True, True)
except Exception,c:
self.assertEqual(True,False)
def test3_optimizer_calculate_gradients(self,mock_scalar,mock_shared, mock_grad):
mock_grad.return_value = 1
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt.calculate_gradients(params= self.input_params, objective=self.input_objective, verbose = self.verbose)
self.assertTrue(numpy.allclose(self.opt.gradients,self.input_params))
def test2_optimizer_init_momentum(self,mock_scalar,mock_shared, mock_ifelse):
mock_ifelse.return_value = self.sample
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt._init_momentum()
self.assertEqual(self.opt.momentum,self.scalar_value)
def test13_optimizer_no_init(self,mock_scalar,mock_shared):
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
optimizer_params = {
}
self.opt = opt(
optimizer_init_args = optimizer_params,
verbose = self.verbose)
self.assertEqual(self.opt.id,'-1')
def test4_optimizer_calculate_gradients_exception(self,mock_scalar,mock_shared):
try:
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt.calculate_gradients(params= self.input_params, objective=False, verbose = self.verbose)
except Exception, c:
self.assertEqual(c.message, self.gradiant_exception_msg)
def test1_optimizer(self,mock_scalar,mock_shared):
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.assertEqual(self.opt.id,self.optimizer_id)
self.assertEqual(self.opt.momentum_start,self.momentum_params[0])
self.assertEqual(self.opt.momentum_end, self.momentum_params[1])
self.assertEqual(self.opt.momentum_epoch_end, self.momentum_params[2])
self.assertEqual(self.opt.momentum_type,self.momentum_type)
self.assertEqual(self.opt.epoch,self.scalar_value)
self.assertEqual(self.opt.learning_rate,self.learning_val)
def test11_optimizer_constrain(self,mock_shared):
A = theano.shared(numpy.asarray([1, 1, 1], dtype=theano.config.floatX))
B = theano.shared(numpy.asarray([1, 1, 1], dtype=theano.config.floatX))
mock_shared.return_value = self.learning_val
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt.params = [A, B]
self.opt.updates = self.f(A) + self.f(B)
self.opt.gradients = [0,0]
self.opt.learning_rate = 0
self.opt._constrain(verbose=self.verbose)
self.assertTrue(numpy.allclose(self.opt.updates[0],A))
def test5_optimizer_nesterov(self,mock_scalar, mock_shared,mock_ifelse):#mock_shared
A = theano.shared(numpy.asarray([1, 1, 1], dtype=theano.config.floatX))
B = theano.shared(numpy.asarray([1, 1, 1], dtype=theano.config.floatX))
mock_ifelse.return_value = 0.9
mock_shared.return_value = self.learning_val
mock_scalar.return_value = self.scalar_value
self.opt = opt(
optimizer_init_args = self.optimizer_params,
verbose = self.verbose)
self.opt._init_momentum()
self.opt.params = [A, B]
self.opt.updates = self.f(A) + self.f(B)
self.opt._nesterov()
self.assertTrue(numpy.allclose(self.opt.updates[0],A))
