# neural networks work better if classes are encoded using
# one output neuron per class
trndata._convertToOneOfMany( bounds=[0,1] )
# same for the independent test data set
tstdata = generateClassificationData(100)
tstdata._convertToOneOfMany( bounds=[0,1] )
# build a feed-forward network with 20 hidden units, plus
# a corresponding trainer
fnn = buildNetwork( trndata.indim, 5, trndata.outdim, outclass=SoftmaxLayer )
trainer = BackpropTrainer( fnn, dataset=trndata, momentum=0.1, verbose=True, weightdecay=0.01)
# generate a grid of data points for visualization
griddata, X, Y = generateGridData([-3.,6.,0.2],[-3.,6.,0.2])
# repeat 20 times
for i in range(20):
# train the network for 1 epoch
trainer.trainEpochs( 1 )
# evaluate the result on the training and test data
trnresult = percentError( trainer.testOnClassData(),
trndata['class'] )
tstresult = percentError( trainer.testOnClassData(
dataset=tstdata ), tstdata['class'] )
# print the result
print("epoch: %4d" % trainer.totalepochs, \
" train error: %5.2f%%" % trnresult, \
# neural networks work better if classes are encoded using
# one output neuron per class
trndata._convertToOneOfMany( bounds=[0,1] )
# same for the independent test data set
tstdata = generateClassificationData(100)
tstdata._convertToOneOfMany( bounds=[0,1] )
# build a feed-forward network with 20 hidden units, plus
# a corresponding trainer
fnn = buildNetwork( trndata.indim, 5, trndata.outdim, outclass=SoftmaxLayer )
trainer = BackpropTrainer( fnn, dataset=trndata, momentum=0.1, verbose=True, weightdecay=0.01)
# generate a grid of data points for visualization
griddata, X, Y = generateGridData([-3.,6.,0.2],[-3.,6.,0.2])
# repeat 20 times
for i in range(20):
# train the network for 1 epoch
trainer.trainEpochs( 1 )
# evaluate the result on the training and test data
trnresult = percentError( trainer.testOnClassData(),
trndata['class'] )
tstresult = percentError( trainer.testOnClassData(
dataset=tstdata ), tstdata['class'] )
# print the result
print "epoch: %4d" % trainer.totalepochs, \
" train error: %5.2f%%" % trnresult, \
svm = SVMUnit()
trainer = SVMTrainer(svm, trndata)
# train the with fixed meta-parameters
log2C = 0. # degree of slack
log2g = 1.1 # width of RBF kernels
trainer.train(log2C=log2C, log2g=log2g)
# alternatively, could train the SVM using design-of-experiments grid search
##trainer.train( search="GridSearchDOE" )
# pass data sets through the SVM to get performance
trnresult = percentError(svm.activateOnDataset(trndata), trndata['target'])
tstresult = percentError(svm.activateOnDataset(tstdata), tstdata['target'])
print "sigma: %7g, C: %7g, train error: %5.2f%%, test error: %5.2f%%" % (
2.0**log2g, 2.0**log2C, trnresult, tstresult)
# generate a grid dataset
griddat, X, Y = generateGridData(x=[-4, 8, 0.1], y=[-2, 3, 0.1])
# pass the grid through the SVM, but this time get the raw distance
# from the boundary, not the class
Z = svm.activateOnDataset(griddat, values=True)
# the output format is a bit weird... make it into a decent array
Z = p.array([z.values()[0] for z in Z]).reshape(X.shape)
# make a 2d plot of training data with an decision value contour overlay
fig = p.figure()
plotData(trndata)
p.contourf(X, Y, Z)
p.show()
log2g = 1.1 # width of RBF kernels
trainer.train(log2C=log2C, log2g=log2g)
# alternatively, could train the SVM using design-of-experiments grid search
##trainer.train( search="GridSearchDOE" )
# pass data sets through the SVM to get performance
trnresult = percentError(svm.activateOnDataset(trndata), trndata["target"])
tstresult = percentError(svm.activateOnDataset(tstdata), tstdata["target"])
print "sigma: %7g, C: %7g, train error: %5.2f%%, test error: %5.2f%%" % (
2.0 ** log2g,
2.0 ** log2C,
trnresult,
tstresult,
)
# generate a grid dataset
griddat, X, Y = generateGridData(x=[-4, 8, 0.1], y=[-2, 3, 0.1])
# pass the grid through the SVM, but this time get the raw distance
# from the boundary, not the class
Z = svm.activateOnDataset(griddat, values=True)
# the output format is a bit weird... make it into a decent array
Z = p.array([z.values()[0] for z in Z]).reshape(X.shape)
# make a 2d plot of training data with an decision value contour overlay
fig = p.figure()
plotData(trndata)
p.contourf(X, Y, Z)
p.show()
# Plot output
#----------
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import cm
from matplotlib.ticker import LinearLocator, FormatStrFormatter
import matplotlib.pyplot as plt
import numpy as np
fig = plt.figure()
ax = fig.gca(projection='3d')
X = np.arange(-2, 2.5, 0.1)
Y = np.arange(-2, 2.5, 0.1)
X, Y = np.meshgrid(X, Y)
# Need to get activation here
griddata, X, Y = generateGridData([-2.,2.5,0.2],[-2.,2.5,0.2])
out = net.activateOnDataset(griddata)
out = out.reshape(X.shape)
# R = np.sqrt(X**2 + Y**2)
# Z = np.sin(R)
surf = ax.plot_surface(X, Y, out, rstride=1, cstride=1, cmap=cm.coolwarm,
linewidth=0, antialiased=False)
ax.set_zlim(-1.01, 1.01)
ax.zaxis.set_major_locator(LinearLocator(10))
ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
fig.colorbar(surf, shrink=0.5, aspect=5)
# Plot output
#----------
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import cm
from matplotlib.ticker import LinearLocator, FormatStrFormatter
import matplotlib.pyplot as plt
import numpy as np
fig = plt.figure()
ax = fig.gca(projection='3d')
X = np.arange(-2, 2.5, 0.1)
Y = np.arange(-2, 2.5, 0.1)
X, Y = np.meshgrid(X, Y)
# Need to get activation here
griddata, X, Y = generateGridData([-2., 2.5, 0.2], [-2., 2.5, 0.2])
out = net.activateOnDataset(griddata)
out = out.reshape(X.shape)
# R = np.sqrt(X**2 + Y**2)
# Z = np.sin(R)
surf = ax.plot_surface(X,
Y,
out,
rstride=1,
cstride=1,
cmap=cm.coolwarm,
linewidth=0,
antialiased=False)
ax.set_zlim(-1.01, 1.01)