def __init__(self,
input_dim,
target_tensor=2,
clip_value=None,
input_tensor=None):
self.target_tensor = dim_to_var(
target_tensor,
"k") if type(target_tensor) == int else target_tensor
self.clip_value = clip_value
super(NeuralRegressor, self).__init__(input_dim,
input_tensor=input_tensor)
def optimize_function(params, config=None):
"""
Create a optimizing function receives gradients.
Parameters:
params - parameters
config - training configuration
Returns:
updating function receives gradients
"""
gs = [dim_to_var(p.ndim) for p in params]
updates, _ = optimize_updates(params, gs, config)
return theano.function(gs, [], updates=updates)
def setup_variables(self):
"""
Set up variables.
"""
if self.input_tensor:
if type(self.input_tensor) == int:
x = dim_to_var(self.input_tensor, name="x")
else:
x = self.input_tensor
else:
x = T.matrix('x')
self.input_variables.append(x)
self._output = x
self._test_output = x
def setup_variables(self):
"""
Set up variables.
"""
if self.input_tensor:
if type(self.input_tensor) == int:
x = dim_to_var(self.input_tensor, name="x")
else:
x = self.input_tensor
else:
x = T.matrix('x')
self.input_variables.append(x)
self._output = x
self._test_output = x
def __init__(self, input_dim, target_tensor=2, clip_value=None, input_tensor=None):
self.target_tensor = dim_to_var(target_tensor, "k") if type(target_tensor) == int else target_tensor
self.clip_value = clip_value
super(NeuralRegressor, self).__init__(input_dim, input_tensor=input_tensor)
