class TanhLayer(Layer):
""" Layer with `tanh` activation function.
Parameters
----------
function_coef : dict
Default configurations for sigmoid activation function. There is one
value name ``alpha`` (default is ``1``). `alpha` control your
function shape.
{layer_params}
"""
function_coef = DictProperty(default={'alpha': 1})
activation_function = tanh
class TanhLayer(Layer):
""" The layer with the `tanh` activation function.
Parameters
----------
function_coef : dict
The default configurations for the sigmoid activation function.
There is one available parameter ``alpha`` (defaults to ``1``).
Parameter `alpha` controls function shape.
{layer_params}
"""
function_coef = DictProperty(default={'alpha': 1})
activation_function = tanh
class SoftmaxLayer(Layer):
""" Layer with softmax activation function.
Parameters
----------
function_coef : dict
Default configurations for softmax activation function. There is one
value name ``temp`` (default is ``1``). Smaller ``temp`` value will
make your winner probability closer to ``1``. To big ``temp`` value
will make all your probabilities closer to equal values.
{layer_params}
"""
function_coef = DictProperty(default={'temp': 1})
activation_function = softmax
class SoftmaxLayer(Layer):
""" The layer with the softmax activation function.
Parameters
----------
function_coef : dict
The default configurations for the softmax activation function.
There is one available parameter ``temp`` (defaults to ``1``).
Lower ``temp`` value will make your winner probability closer
to ``1``. Higher ``temp`` value will make all probabilities
values equal to each other.
{layer_params}
"""
function_coef = DictProperty(default={'temp': 1})
activation_function = softmax
class Layer(BaseLayer):
""" Base class for neural network layers.
Parameters
----------
function_coef : dict
Default settings for activation function.
{layer_params}
"""
function_coef = DictProperty()
def __init__(self, *args, **kwargs):
super(Layer, self).__init__(*args, **kwargs)
if self.function_coef is not None:
partial_func = get_partial_for_func(self.activation_function)
self.activation_function = partial_func(self.activation_function,
**self.function_coef)