def __init__(self, ls_neurons_per_layers, used_sectors, learning_rate):
''' Initialization ... same parameters as for DiscreteMLMVN
@see DiscreteMLMVN.__init__
'''
## Sectors used for counting activation function of neurons in last
# layer
self.sects = used_sectors
## Learning rate defining speed of learning
self.learning_rate = learning_rate
## List of layers of given network
self.ls_layers = []
# create first and hidden layers
for (n_inputs, n_neurons) in zip(ls_neurons_per_layers[:-2],
ls_neurons_per_layers[1:-1]):
self.ls_layers.append(
layer.MVNLayer(n_neurons,
n_inputs,
learning_rate=self.learning_rate))
# create last layer
n_inputs = ls_neurons_per_layers[-2]
n_neurons = ls_neurons_per_layers[-1]
self.ls_layers.append(
layer.MVNLastLayer(n_neurons,
n_inputs,
activation_func=self.sects.activation_function,
learning_rate=self.learning_rate))
# set upper layers of self.ls_layers for backpropagation alg.
for (this_l, upper_l) in zip(self.ls_layers[:-1], self.ls_layers[1:]):
this_l.set_upper_layer(upper_l)
def __init__(self, ls_neurons_per_layers, learning_rate=1):
''' Initialization ... same parameters as for DiscreteMLMVN, but
we do not need info about sectors of neurons.
@see DiscreteMLMVN.__init__
'''
## Learning rate defining speed of learning
self.learning_rate = learning_rate
## List of layers of given network
self.ls_layers = []
# create first and hidden layers wth continuous neurons
for (n_inputs, n_neurons) in zip(ls_neurons_per_layers[:-2],
ls_neurons_per_layers[1:-1]):
self.ls_layers.append(
layer.MVNLayer(n_neurons,
n_inputs,
learning_rate=self.learning_rate))
# create last layer wth continuous neurons
n_inputs = ls_neurons_per_layers[-2]
n_neurons = ls_neurons_per_layers[-1]
self.ls_layers.append(
layer.MVNLastLayer(n_neurons,
n_inputs,
learning_rate=self.learning_rate))
# set upper layers of self.ls_layers for backpropagation alg.
for (this_l, upper_l) in zip(self.ls_layers[:-1], self.ls_layers[1:]):
this_l.set_upper_layer(upper_l)
def __init__(self, ls_neurons_per_layers, used_sectors, learning_rate):
''' Initializes network
@param ls_neurons_per_layers List specifying number of layers and
neurons in them.\n Example: [<#inputs for first layer>,
<#neurons fst layer>, <#neurons hidden layer>,
<#neurons out layer>]\n
Example 2:
@verbatim
[20,8,4,2] therefore indicates:
layer 1: 20 inputs, 8 neurons
layer 2: 8 inputs, 4 outputs
last layer: 4 inputs, 2 outputs (network outputs)
@endverbatim
@param layers_list List of [(n_inputs,n_neurons) .. ]
@param used_sectors Sectors used to encode/transform data.\n
Sectors provide activation function for neurons.\n
If ommited, continuous sectors/activation function
is used through whole network.
@param learning_rate Specifies speed of learning should be in
interval (0,1] (where 1 == full speed), but higher
speed is also possible.
'''
## Sectors used for counting activation function of neurons
self.sects = used_sectors
## Learning rate defining speed of learning
self.learning_rate = learning_rate
## List of layers of given network
self.ls_layers = []
# create first and hidden layers
for (n_inputs, n_neurons) in zip(ls_neurons_per_layers[:-2],
ls_neurons_per_layers[1:-1]):
self.ls_layers.append(
layer.MVNLayer(n_neurons,
n_inputs,
activation_func=self.sects.activation_function,
learning_rate=self.learning_rate))
# create last layer
n_inputs = ls_neurons_per_layers[-2]
n_neurons = ls_neurons_per_layers[-1]
self.ls_layers.append(
layer.MVNLastLayer(n_neurons,
n_inputs,
activation_func=self.sects.activation_function,
learning_rate=self.learning_rate))
# set upper layers of self.ls_layers for backpropagation alg.
for (this_l, upper_l) in zip(self.ls_layers[:-1], self.ls_layers[1:]):
this_l.set_upper_layer(upper_l)
def __init__(self, ls_neurons_per_layers, learning_rate):
''' Initializes network
@param ls_neurons_per_layers List specifying number of layers and
neurons in them.\n Example: [<#inputs for first layer>,
<#neurons fst layer>, <#neurons hidden layer>,
<#neurons out layer>]\n
Example 2:
@verbatim
[20,8,4,2] therefore indicates:
layer 1: 20 inputs, 8 neurons
layer 2: 8 inputs, 4 outputs
last layer: 4 inputs, 2 outputs (network outputs)
@endverbatim
@param layers_list List of [(n_inputs,n_neurons) .. ]
@param learning_rate Specifies speed of learning should be in
interval (0,1] (where 1 == full speed), but higher
speed is also possible.
'''
## List of layers of given network
self.ls_layers = []
# create first and hidden layers with continuous neurons
for (n_inputs, n_neurons) in zip(ls_neurons_per_layers[:-2],
ls_neurons_per_layers[1:-1]):
self.ls_layers.append(layer.MVNLayer(n_neurons, n_inputs))
# create last layer with continuous neurons
n_inputs = ls_neurons_per_layers[-2]
n_neurons = ls_neurons_per_layers[-1]
self.ls_layers.append(layer.MVNLastLayer(n_neurons, n_inputs))
# set upper layers of self.ls_layers for backpropagation alg.
for (this_l, upper_l) in zip(self.ls_layers[:-1], self.ls_layers[1:]):
this_l.set_upper_layer(upper_l)
# set learning rate for each layer
self.set_learning_rate(learning_rate)