def gda_model_accuracy_nmf(Xraw, L, num_features, percent_train=0.5):
# reduce dimensions of raw data using NMF
nmf_obj = skd.NMF(n_components=num_features, random_state=17)
X = nmf_obj.fit_transform(Xraw.T).T
Q = nmf_obj.components_.T
# get shape information for labels
(nl, ) = L.shape
# break data into training and testing datasets
(Dtr, Dtt) = mydata.separateClassData(
X, L.reshape(1, nl), numdata_or_percent_for_training=percent_train)
# get the start time the training starts
start = time.time()
# build the set of LDA models
dlist = mydsc.constructDistriminantSet(X=Dtr['net'], L=Dtr['nlbl'])
# get the end time after training
end = time.time()
print('({0}) Time elapsed to train is: '.format(num_features), end - start)
# test the discriminants for each label dataset in the training set
Ltest = mydsc.evalDiscriminantSet(X=Dtt['net'], discriminant_list=dlist)
# Compute accuracy and confusion matrix
classes = np.array([0, 1])
(C, accuracy) = mydsc.computeConfusionMatrix(Xtest=Dtt['net'],
Ltest=Dtt['nlbl'],
Leval=Ltest,
classes=classes)
# plot the confusion matrix
f0 = plot.figure()
(f0, ax0) = myimg.plotDataset(f0, C, delta=(-1, -1))
ax0.set_xlabel('Classified Classes')
ax0.set_ylabel('True Classes')
f0.savefig('{0}/confusion_mat_{1}.png'.format('../plots/GDA',
num_features))
# print output message
print('Accuracy for {0} features is : '.format(num_features), accuracy)
# return the accuracy
return accuracy / 100.0
# evaluate the discriminant functions
Labels = mydisc.evalDiscriminantSet(Test[vars], set)
# find time the training finished
end = timer()
print('Testing took {0} seconds'.format(end - start))
# compute the confusion matrix along with accuracy
(C, accuracy) = mydisc.computeConfusionMatrix(Test[vars], Test[lbls], Labels,
classes)
print('The overall accuracy is {0}%'.format(accuracy))
# plot the confusion matrix
plotdir = 'plots/prob1_face'
f0 = plot.figure()
(f0, ax0) = myimg.plotDataset(f0, C, delta=(-1, -1))
ax0.set_xlabel('Classified Classes')
ax0.set_ylabel('True Classes')
f0.savefig('{0}/confusion_mat.png'.format(plotdir))
# Find the largest confusion values in the matrix and get their tuple pair
cbest = mydisc.computeMostConfused(C, num_values=4)
# For each pair of confused values, generate and save each digit's likelihood and the
# pair's log odds ratio
for (i1, i2, cval) in cbest:
fig = plot.figure()
l1 = set[i1].likelihood(class_value=1).reshape(35, 30)
l2 = set[i2].likelihood(class_value=1).reshape(35, 30)
odds = l1 - l2
# compute the confusion matrix along with accuracy
(C,accuracy) = mydisc.computeConfusionMatrix(Test[vars],Test[lbls],Labels,classes)
print('({0},{1})| The overall accuracy is {2}%'.format(dim[0], dim[1], accuracy))
# generate appropriate paths
plotdir0 = 'plots/prob12_disjoint'
plotdir = 'plots/prob12_disjoint/{0}_{1}'.format(dim[0],dim[1])
try:
os.makedirs(plotdir0)
except OSError as e:
if e.errno != errno.EEXIST:
raise
try:
os.makedirs(plotdir)
except OSError as e:
if e.errno != errno.EEXIST:
raise
# plot the confusion matrix
f0 = plot.figure()
(f0,ax0) = myimg.plotDataset(f0,C,delta=(-1,-1))
ax0.set_xlabel('Classified Classes')
ax0.set_ylabel('True Classes')
f0.savefig('{0}/confusion_mat.png'.format(plotdir))
# index for independent var data and labels
vars = 0
lbls = 1
# define classes
classes = np.array([1, 2, 3, 4, 5])
# load the dataset in a proper form
(Test,Train) = myload.loadProblem2Part2('data/prob2/part2')
# create the discriminant functions
possible_values = np.array([0,1])
set = mydisc.constructDistriminantSet(X=Train[vars],
L=Train[lbls],
possible_values=possible_values,
laplace_smoothing_val=1e-1)
# evaluate the discriminant functions
Labels = mydisc.evalDiscriminantSet(Test[vars],set,classes=classes)
# compute the confusion matrix along with accuracy
(C,accuracy) = mydisc.computeConfusionMatrix(Test[vars],Test[lbls],Labels,classes)
print('The overall accuracy is {0}%'.format(accuracy))
# plot the confusion matrix
plotdir = 'plots/prob22'
f0 = plot.figure()
(f0,ax0) = myimg.plotDataset(f0,C,delta=(0,0))
ax0.set_xlabel('Classified Classes')
ax0.set_ylabel('True Classes')
f0.savefig('{0}/confusion_mat.png'.format(plotdir))