def testPlot1():
dim = 15
from scipy import rand, dot
from pybrain.datasets import SupervisedDataSet
from pybrain import LinearLayer, FullConnection, FeedForwardNetwork
from pybrain.utilities import dense_orth
net = FeedForwardNetwork()
net.addInputModule(LinearLayer(dim, name='in'))
net.addOutputModule(LinearLayer(1, name='out'))
net.addConnection(FullConnection(net['in'], net['out']))
net.sortModules()
ds = SupervisedDataSet(dim, 1)
ds2 = SupervisedDataSet(dim, 1)
R = dense_orth(dim)
for _ in range(1000):
tmp = rand(dim) > 0.5
tmp2 = dot(tmp, R)
ds.addSample(tmp, [tmp[-1]])
ds2.addSample(tmp2, [tmp[-1]])
f = ModuleWrapper(ds, net)
f2 = ModuleWrapper(ds2, net)
# tracking progress by callback
ltrace = []
def storer(a):
ltrace.append(a.provider.currentLosses(a.bestParameters))
x = net.params
x *= 0.001
algo = SGD(f, net.params.copy(), callback=storer, learning_rate=0.2)
algo.run(1000)
pylab.plot(ltrace, 'r-')
del ltrace[:]
algo = SGD(f2, net.params.copy(), callback=storer, learning_rate=0.2)
algo.run(1000)
pylab.plot(ltrace, 'g-')
pylab.semilogy()
pylab.show()
def testPlot1():
dim = 15
from scipy import rand, dot
from pybrain.datasets import SupervisedDataSet
from pybrain import LinearLayer, FullConnection, FeedForwardNetwork
from pybrain.utilities import dense_orth
net = FeedForwardNetwork()
net.addInputModule(LinearLayer(dim, name='in'))
net.addOutputModule(LinearLayer(1, name='out'))
net.addConnection(FullConnection(net['in'], net['out']))
net.sortModules()
ds = SupervisedDataSet(dim, 1)
ds2 = SupervisedDataSet(dim, 1)
R = dense_orth(dim)
for _ in range(1000):
tmp = rand(dim) > 0.5
tmp2 = dot(tmp, R)
ds.addSample(tmp, [tmp[-1]])
ds2.addSample(tmp2, [tmp[-1]])
f = ModuleWrapper(ds, net)
f2 = ModuleWrapper(ds2, net)
# tracking progress by callback
ltrace = []
def storer(a):
ltrace.append(a.provider.currentLosses(a.bestParameters))
x = net.params
x *= 0.001
algo = SGD(f, net.params.copy(), callback=storer, learning_rate=0.2)
algo.run(1000)
pylab.plot(ltrace, 'r-')
del ltrace[:]
algo = SGD(f2, net.params.copy(), callback=storer, learning_rate=0.2)
algo.run(1000)
pylab.plot(ltrace, 'g-')
pylab.semilogy()
pylab.show()
def __init__(
self,
basef,
translate=True,
rotate=False,
conditioning=None,
asymmetry=None,
oscillate=False,
penalized=0,
desiredValue=1e-8,
gnoise=None,
unoise=None,
cnoise=None,
sparse=True,
):
FunctionEnvironment.__init__(self, basef.xdim, basef.xopt)
self._name = basef.__class__.__name__
self.desiredValue = desiredValue
self.toBeMinimized = basef.toBeMinimized
if self.xdim < 500:
sparse = False
if sparse:
try:
from scipy.sparse import csc_matrix
except:
sparse = False
if translate:
self.xopt = (rand(self.xdim) - 0.5) * 9.8
if conditioning:
prefix = generateDiags(conditioning, self.xdim)
if sparse:
prefix = csc_matrix(prefix)
if rotate:
prefix = prefix * sparse_orth(self.xdim)
if oscillate or not asymmetry:
prefix = sparse_orth(self.xdim) * prefix
else:
if rotate:
prefix = dot(prefix, dense_orth(self.xdim))
if oscillate or not asymmetry:
prefix = dot(dense_orth(self.xdim), prefix)
elif rotate and asymmetry and not oscillate:
if sparse:
prefix = sparse_orth(self.xdim)
else:
prefix = dense_orth(self.xdim)
elif sparse:
prefix = None
else:
prefix = eye(self.xdim)
if penalized != 0:
if self.penalized:
penalized = 0
else:
self.penalized = True
# combine transformations
if rotate:
if sparse:
r = sparse_orth(self.xdim)
tmp1 = lambda x: ravel(x * r)
else:
r = dense_orth(self.xdim)
tmp1 = lambda x: dot(x, r)
else:
tmp1 = lambda x: x
if oscillate:
tmp2 = lambda x: BBOBTransformationFunction.oscillatify(tmp1(x))
else:
tmp2 = tmp1
if asymmetry is not None:
tmp3 = lambda x: BBOBTransformationFunction.asymmetrify(
tmp2(x), asymmetry)
else:
tmp3 = tmp2
# noise
ntmp = None
if gnoise:
ntmp = lambda f: f * exp(gnoise * gauss(0, 1))
elif unoise:
alpha = 0.49 * (1. / self.xdim) * unoise
ntmp = lambda f: f * power(random(), unoise) * max(
1, power(1e9 / (f + 1e-99), alpha * random()))
elif cnoise:
alpha, beta = cnoise
ntmp = lambda f: f + alpha * max(
0, 1000 * (random() < beta) * gauss(0, 1) /
(abs(gauss(0, 1)) + 1e-199))
def noisetrans(f):
if ntmp is None or f < 1e-8:
return f
else:
return ntmp(f) + 1.01e-8
if sparse:
if prefix is None:
tmp4 = lambda x: tmp3(x - self.xopt)
else:
tmp4 = lambda x: ravel(prefix * tmp3(x - self.xopt))
else:
tmp4 = lambda x: dot(prefix, tmp3(x - self.xopt))
self.f = lambda x: (noisetrans(basef.f(tmp4(x))) + penalized *
penalize(x))
def __init__(self, basef,
translate=True,
rotate=False,
conditioning=None,
asymmetry=None,
oscillate=False,
penalized=0,
desiredValue=1e-8,
gnoise=None,
unoise=None,
cnoise=None,
sparse=True,
):
FunctionEnvironment.__init__(self, basef.xdim, basef.xopt)
self._name = basef.__class__.__name__
self.desiredValue = desiredValue
self.toBeMinimized = basef.toBeMinimized
if self.xdim < 500:
sparse = False
if sparse:
try:
from scipy.sparse import csc_matrix
except:
sparse = False
if translate:
self.xopt = (rand(self.xdim) - 0.5) * 9.8
if conditioning:
prefix = generateDiags(conditioning, self.xdim)
if sparse:
prefix = csc_matrix(prefix)
if rotate:
prefix = prefix * sparse_orth(self.xdim)
if oscillate or not asymmetry:
prefix = sparse_orth(self.xdim) * prefix
else:
if rotate:
prefix = dot(prefix, dense_orth(self.xdim))
if oscillate or not asymmetry:
prefix = dot(dense_orth(self.xdim), prefix)
elif rotate and asymmetry and not oscillate:
if sparse:
prefix = sparse_orth(self.xdim)
else:
prefix = dense_orth(self.xdim)
elif sparse:
prefix = None
else:
prefix = eye(self.xdim)
if penalized != 0:
if self.penalized:
penalized = 0
else:
self.penalized = True
# combine transformations
if rotate:
if sparse:
r = sparse_orth(self.xdim)
tmp1 = lambda x: ravel(x * r)
else:
r = dense_orth(self.xdim)
tmp1 = lambda x: dot(x, r)
else:
tmp1 = lambda x: x
if oscillate:
tmp2 = lambda x: BBOBTransformationFunction.oscillatify(tmp1(x))
else:
tmp2 = tmp1
if asymmetry is not None:
tmp3 = lambda x: BBOBTransformationFunction.asymmetrify(tmp2(x), asymmetry)
else:
tmp3 = tmp2
# noise
ntmp = None
if gnoise:
ntmp = lambda f: f * exp(gnoise * gauss(0, 1))
elif unoise:
alpha = 0.49 * (1. / self.xdim) * unoise
ntmp = lambda f: f * power(random(), unoise) * max(1, power(1e9 / (f + 1e-99), alpha * random()))
elif cnoise:
alpha, beta = cnoise
ntmp = lambda f: f + alpha * max(0, 1000 * (random() < beta) * gauss(0, 1) / (abs(gauss(0, 1)) + 1e-199))
def noisetrans(f):
if ntmp is None or f < 1e-8:
return f
else:
return ntmp(f) + 1.01e-8
if sparse:
if prefix is None:
tmp4 = lambda x: tmp3(x - self.xopt)
else:
tmp4 = lambda x: ravel(prefix * tmp3(x - self.xopt))
else:
tmp4 = lambda x: dot(prefix, tmp3(x - self.xopt))
self.f = lambda x: (noisetrans(basef.f(tmp4(x)))
+ penalized * penalize(x))