def train_momentum(self, input, target, eta = 0.2, momentum = 0.8, \
maxiter = 10000, disp = 0):
"""
Simple backpropagation training with momentum.
:Parameters:
input : 2-D array
Array of input patterns
target : 2-D array
Array of network targets
eta : float, optional
Learning rate
momentum : float, optional
Momentum coefficient
maxiter : integer, optional
Maximum number of iterations
disp : bool
If True convergence method is displayed
"""
input, target = self._setnorm(input, target)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" %err
self.weights = netprop.momentum(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, eta, momentum, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" %err
def train_rprop(self, input, target, \
a = 1.2, b = 0.5, mimin = 0.000001, mimax = 50., \
xmi = 0.1, maxiter = 10000, disp = 0):
"""
Rprop training algorithm.
:Parameters:
input : 2-D array
Array of input patterns
target : 2-D array
Array of network targets
a : float, optional
Training step increasing parameter
b : float, optional
Training step decreasing parameter
mimin : float, optional
Minimum training step
mimax : float, optional
Maximum training step
xmi : array (or float), optional
Array containing initial training steps for weights.
If *xmi* is a scalar then its value is set for all weights
maxiter : integer, optional
Maximum number of iterations
disp : bool
If True convergence method is displayed. Default is *False*
:Returns:
xmi : array
Computed array of training steps to be used in eventual further
training calls.
"""
input, target = self._setnorm(input, target)
if type(xmi).__name__ in ['float', 'int']:
xmi = [ xmi ]*len(self.conec)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" %err
self.weights, xmi = netprop.rprop(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, a, b, mimin, mimax, xmi, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" %err
return xmi
def sqerror(self, input, target):
"""
Calculates sum of squared errors at network output.
Error is calculated for **normalized** input and target arrays.
:Parameters:
input : 2-D array
Array of input patterns
target : 2-D array
Array of network targets
:Returns:
err : float
0.5*(sum of squared errors at network outputs)
.. note::
This function might be slow in frequent use, because data
normalization is performed at each call. Usually there's no need
to use this function, unless you need to adopt your own training
strategy.
"""
input, target = self._setnorm(input, target)
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
return err
def sqerror(self, input, target):
"""
Returns 0.5*(sum of squared errors at output)
for input and target arrays being first normalized.
Might be slow in frequent use, because data normalization is
performed at each call.
"""
input, target = self._setnorm(input, target)
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
return err
def sqerror(self, input, target):
"""
Returns 0.5*(sum of squared errors at output)
for input and target arrays being first normalized.
Might be slow in frequent use, because data normalization is
performed at each call.
"""
input, target = self._setnorm(input, target)
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
return err
def train_momentum(self, input, target, eta = 0.2, momentum = 0.8, \
maxiter = 10000, disp = 0):
"""
Simple backpropagation training with momentum.
Allowed parameters:
eta - descent scaling parameter (default is 0.2)
momentum - momentum coefficient (default is 0.8)
maxiter - the maximum number of iterations (default is 10000)
disp - print convergence message if non-zero (default is 0)
"""
input, target = self._setnorm(input, target)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" %err
self.weights = netprop.momentum(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, eta, momentum, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" %err
def train_momentum(self, input, target, eta = 0.2, momentum = 0.8, \
maxiter = 10000, disp = 0):
"""
Simple backpropagation training with momentum.
Allowed parameters:
eta - descent scaling parameter (default is 0.2)
momentum - momentum coefficient (default is 0.8)
maxiter - the maximum number of iterations (default is 10000)
disp - print convergence message if non-zero (default is 0)
"""
input, target = self._setnorm(input, target)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" % err
self.weights = netprop.momentum(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, eta, momentum, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" % err
def train_rprop(self, input, target, \
a = 1.2, b = 0.5, mimin = 0.000001, mimax = 50., \
xmi = 0.1, maxiter = 10000, disp = 0):
"""
Rprop training algorithm.
Allowed parameters:
a - training step increasing parameter (default is 1.2)
b - training step decreasing parameter (default is 0.5)
mimin - minimum training step (default is 0.000001)
mimax - maximum training step (default is 50.)
xmi - vector containing initial training steps for weights;
if 'xmi' is a scalar then its value is set for all
weights (default is 0.1)
maxiter - the maximum number of iterations (default is 10000)
disp - print convergence message if non-zero (default is 0)
Method updates network weights and returns 'xmi' vector
(after 'maxiter' iterations).
"""
input, target = self._setnorm(input, target)
if type(xmi).__name__ in ['float', 'int']:
xmi = [ xmi ]*len(self.conec)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" %err
self.weights, xmi = netprop.rprop(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, a, b, mimin, mimax, xmi, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" %err
return xmi
def train_rprop(self, input, target, \
a = 1.2, b = 0.5, mimin = 0.000001, mimax = 50., \
xmi = 0.1, maxiter = 10000, disp = 0):
"""
Rprop training algorithm.
Allowed parameters:
a - training step increasing parameter (default is 1.2)
b - training step decreasing parameter (default is 0.5)
mimin - minimum training step (default is 0.000001)
mimax - maximum training step (default is 50.)
xmi - vector containing initial training steps for weights;
if 'xmi' is a scalar then its value is set for all
weights (default is 0.1)
maxiter - the maximum number of iterations (default is 10000)
disp - print convergence message if non-zero (default is 0)
Method updates network weights and returns 'xmi' vector
(after 'maxiter' iterations).
"""
input, target = self._setnorm(input, target)
if type(xmi).__name__ in ['float', 'int']:
xmi = [xmi] * len(self.conec)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Initial error --> 0.5*(sum of squared errors at output): %.15f" % err
self.weights, xmi = netprop.rprop(self.weights, self.conec, self.bconecno, \
self.units, self.inno, self.outno, input, \
target, a, b, mimin, mimax, xmi, maxiter)
if disp:
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
print "Final error --> 0.5*(sum of squared errors at output): %.15f" % err
return xmi
def sqerror(self, input, target):
"""
Returns 0.5*(sum of squared errors at output)
for input and target arrays being first normalized.
Might be slow in frequent use, because data normalization is
performed at each call.
Warning:
_setnorm should be called before sqerror - will be changed in future.
"""
input, target = self._testdata(input, target)
input = _normarray(input, self.eni) #Normalization data might be uninitialized here!
target = _normarray(target, self.eno)
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
return err
def sqerror(self, input, target):
"""
Returns 0.5*(sum of squared errors at output)
for input and target arrays being first normalized.
Might be slow in frequent use, because data normalization is
performed at each call.
Warning:
_setnorm should be called before sqerror - will be changed in future.
"""
input, target = self._testdata(input, target)
input = _normarray(
input, self.eni) #Normalization data might be uninitialized here!
target = _normarray(target, self.eno)
err = netprop.sqerror(self.weights, self.conec, self.units, \
self.inno, self.outno, input, target)
return err