# Fit and plot one-vs-rest NN classifiers
y_test_est = np.mat(np.zeros((y_test.shape[0], C)))
nets = []
for c in range(C):
ann = nl.net.newff(
[[X[:, 0].min(), X[:, 0].max()], [X[:, 1].min(), X[:, 1].max()]],
[NHiddenUnits, 1],
[nl.trans.TanSig(), nl.trans.PureLin()])
train_error = ann.train(X_train, (y_train == c),
goal=learning_goal,
epochs=max_epochs,
show=round(max_epochs / 8))
nets.append(ann)
y_test_est[:, c] = ann.sim(X_test)
figure(c + 1)
dbplotf(X_test, (y_test == c).astype(int), lambda x: ann.sim(x), 'auto')
# Plot descision boundary for ensemble of neural networks
figure(C + 1)
def neval(xval):
return np.argmax(hstack([n.sim(xval) for n in nets]), 1)
dbplotf(X_test, y_test, neval, 'auto')
show()
# Compute error rate
ErrorRate = (np.argmax(y_test_est, 1) != y_test).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100 * ErrorRate))
DS_train = conv2DS(X_train, y_train)
DS_test = conv2DS(X_test, y_test)
fnn = buildNetwork(DS_train.indim,
NHiddenUnits,
DS_train.outdim,
outclass=SoftmaxLayer,
bias=True)
trainer = BackpropTrainer(fnn,
dataset=DS_train,
momentum=0.1,
verbose=True,
weightdecay=0.01)
# Train for 100 iterations.
for i in range(50):
trainer.trainEpochs(1)
ote = fnn.activateOnDataset(DS_test)
ErrorRate = (np.argmax(ote, 1) != y_test.T).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100 * ErrorRate))
figure(1)
def neval(xval):
return argmax(fnn.activateOnDataset(conv2DS(np.asmatrix(xval))), 1)
toolbox_02450.dbplotf(X_test, y_test, neval, 'auto')
show()
#%% convert to ClassificationDataSet format.
def conv2DS(Xv,yv = None) :
if yv == None :
yv = np.asmatrix( np.ones( (Xv.shape[0],1) ) )
for j in range(len(classNames)) : yv[j] = j
C = len(unique(yv.flatten().tolist()[0]))
DS = ClassificationDataSet(M, 1, nb_classes=C)
for i in range(Xv.shape[0]) : DS.appendLinked(Xv[i,:].tolist()[0], [yv[i].A[0][0]])
DS._convertToOneOfMany( )
return DS
DS_train = conv2DS(X_train,y_train)
DS_test = conv2DS(X_test,y_test)
# A neural network without a hidden layer will simulate logistic regression (albeit very slowly)
fnn = buildNetwork( DS_train.indim, DS_train.outdim, outclass=SoftmaxLayer,bias=True )
trainer = BackpropTrainer( fnn, dataset=DS_train, momentum=0.1, verbose=True, weightdecay=0.01)
# Train for 100 iterations.
for i in range(50): trainer.trainEpochs( 1 )
ote = fnn.activateOnDataset(DS_test)
ErrorRate = (np.argmax(ote,1) != y_test.T).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100*ErrorRate))
figure(1)
def neval(xval):
return argmax(fnn.activateOnDataset(conv2DS(np.asmatrix(xval)) ),1)
dbplotf(X_test,y_test,neval,'auto')
show()
best_knn = KNeighborsClassifier(n_neighbors=best_param_knn[k])
best_knn = best_knn.fit(X_train, y_train)
y_est = best_knn.predict(X_test)
X_test_outer.append(X_test)
best_knn_list.append(best_knn)
mis_classified = np.sum(y_est != y_test)
relative_error = mis_classified / len(y_test)
y_ESTKNN.append(y_est)
y_test_outer.append(y_test)
err_test_outer.append(relative_error)
k += 1
print("The generalization error is {0} ".format(np.mean(err_test_outer)))
best_index = np.argmin(err_test_outer)
confmatplot(y_test_outer[best_index], y_ESTKNN[best_index])
show()
figure(1)
def neval(xval):
return np.argmax(best_knn.predict_proba(xval), 1)
if k_pca == 2:
figure()
dbplotf(X_test_outer[best_index], y_test_outer[best_index], neval, 'auto')
show()
best_est.append(y_est)
mis_classified = np.sum(y_est != y_test)
relative_error = mis_classified / len(y_test)
gen_err[k] = relative_error
k += 1
# show confusion matrix for model training using K-fold cross validation
best_index = gen_err.argmin()
figure()
confmatplot(best_est[best_index], y_test_outer[best_index])
print("The mean generalization error is {0}".format(np.mean(gen_err)))
#
#
#
# #show confusion matrix for model testing using Leave one out cross validation
# figure(1)
# Y_est2 = clf_list[best_index].predict(X_test)
# confmatplot(Y_test,Y_est2)
# show()
#
# Decision boundaries for the multinomial regression model
def nevallog(xval):
return np.argmax(clf_list[best_index].predict_proba(xval),1)
if k_pca == 2:
figure()
dbplotf(x_test_outer[best_index], y_test_outer[best_index] ,nevallog,'auto')
show()
print y_train.shape
#%% convert to ClassificationDataSet format.
def conv2DS(Xv,yv = None) :
if yv == None :
yv = np.asmatrix( np.ones( (Xv.shape[0],1) ) )
for j in range(len(classNames)) : yv[j] = j
C = len(unique(yv.flatten().tolist()[0]))
DS = ClassificationDataSet(M, 1, nb_classes=C)
for i in range(Xv.shape[0]) : DS.appendLinked(Xv[i,:].tolist()[0], [yv[i].A[0][0]])
DS._convertToOneOfMany( )
return DS
DS_train = conv2DS(X_train,y_train)
DS_test = conv2DS(X_test,y_test)
fnn = buildNetwork(DS_train.indim, NHiddenUnits, DS_train.outdim, outclass=SoftmaxLayer,bias=True)
trainer = BackpropTrainer( fnn, dataset=DS_train, momentum=0.1, verbose=True, weightdecay=0.01)
# Train for 100 iterations.
for i in range(50): trainer.trainEpochs( 1 )
ote = fnn.activateOnDataset(DS_test)
ErrorRate = (np.argmax(ote,1) != y_test.T).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100*ErrorRate))
figure(1)
def neval(xval):
return argmax(fnn.activateOnDataset(conv2DS(np.asmatrix(xval)) ),1)
toolbox_02450.dbplotf(X_test,y_test,neval,'auto')
show()
attributeNames = [name[0] for name in mat_data['attributeNames'].squeeze()]
classNames = [name[0][0] for name in mat_data['classNames']]
N, M = X.shape
C = len(classNames)
NHiddenUnits = 4;
# These parameters are usually adjusted to: (1) data specifics, (2) computational constraints
learning_goal = 0.01 # stop criterion 1 (train mse to be reached)
max_epochs = 300 # stop criterion 2 (max epochs in training)
# Fit and plot one-vs-rest NN classifiers
y_test_est = np.mat(np.zeros((y_test.shape[0],C)))
nets = []
for c in range(C):
ann = nl.net.newff([[X[:,0].min(),X[:,0].max()], [X[:,1].min(),X[:,1].max()]], [NHiddenUnits, 1], [nl.trans.TanSig(),nl.trans.PureLin()])
train_error = ann.train(X_train, (y_train == c), goal=learning_goal, epochs=max_epochs, show=round(max_epochs/8))
nets.append(ann)
y_test_est[:,c] = ann.sim(X_test)
figure(c+1)
dbplotf(X_test,(y_test==c).astype(int), lambda x : ann.sim(x), 'auto')
# Plot descision boundary for ensemble of neural networks
figure(C+1)
def neval(xval):
return np.argmax(hstack([n.sim(xval) for n in nets] ),1)
dbplotf(X_test,y_test,neval,'auto')
show()
# Compute error rate
ErrorRate = (np.argmax(y_test_est,1) != y_test).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100*ErrorRate))
NHiddenUnits = 2;
#%% convert to ClassificationDataSet format.
def conv2DS(Xv,yv = None) :
if yv == None :
yv = np.asmatrix( np.ones( (Xv.shape[0],1) ) )
for j in range(len(classNames)) : yv[j] = j
C = len(unique(yv.flatten().tolist()[0]))
DS = ClassificationDataSet(M, 1, nb_classes=C)
for i in range(Xv.shape[0]) : DS.appendLinked(Xv[i,:].tolist()[0], [yv[i].A[0][0]])
DS._convertToOneOfMany( )
return DS
DS_train = conv2DS(X_train,y_train)
DS_test = conv2DS(X_test,y_test)
fnn = buildNetwork( DS_train.indim, NHiddenUnits, DS_train.outdim, outclass=SoftmaxLayer,bias=True )
trainer = BackpropTrainer( fnn, dataset=DS_train, momentum=0.1, verbose=True, weightdecay=0.01)
# Train for 100 iterations.
for i in range(50): trainer.trainEpochs( 1 )
ote = fnn.activateOnDataset(DS_test)
ErrorRate = (np.argmax(ote,1) != y_test.T).mean(dtype=float)
print('Error rate (ensemble): {0}%'.format(100*ErrorRate))
figure(1)
def neval(xval):
return argmax(fnn.activateOnDataset(conv2DS(np.asmatrix(xval)) ),1)
dbplotf(X_test,y_test,neval,'auto')
show()
