id = np.random.randint(0, nsamp, ntest)
d_test = dat[id, :]
c_test_true = clas[id]
d_train = np.delete(dat, id, 0)
c_train = np.delete(clas, id)
print 'extracting subsample of points for training and classification test'
for i in range(ntest):
print id[i], d_test[i, :]
print ''
print 'fitting k nearest neighbour'
#perform a k nearest neighbour algorithm on the test data to predict which level of heart
#disease will be present in a test patient
c_test = hdp.k_test(d_test, d_train, c_train, distance=2, k=3)
print 'comparing the test with true classifications'
for i in range(ntest):
print 'test class: true class ', c_test[i], c_test_true[i]
idmatch = np.where(c_test == c_test_true)[0]
nmatch = np.shape(idmatch)[0]
print 'accuracy...', 1. * nmatch / ntest
#plot the individual pairs of dimensions in the parameter space to investigate the correlations
if (diagplots == 1):
os.system('rm -rf diagplots')
os.system('mkdir diagplots')
idclas = np.unique(clas)
nclass = np.shape(idclas)[0]
a[:, :] = 1. * X_transformed
XT_train = np.delete(a, itest, 0)
classifier.fit(XT_train[:, :k], trainclas)
# transform new data using already fitted pca
# (don't re-fit the pca)
newdata_transformed = pca.transform(testdat)
# predict labels using the trained classifier
pred_labels = classifier.predict(newdata_transformed[:, :k])
#now use my k nearest neighbour
knn_test = hdp.k_test(newdata_transformed[:, :k],
XT_train[:, :k],
trainclas,
distance=2,
k=3)
print 'pca aided classification'
fracgood = np.shape(np.where(pred_labels == testclas)[0])[0] / 1. / ntest
fracgood_k = np.shape(np.where(knn_test == testclas)[0])[0] / 1. / ntest
for i in range(ntest):
print pred_labels[i], testclas[i], knn_test[i]
print 'frac good', fracgood, fracgood_k
#now without using pca
classifier = DecisionTreeClassifier()
X_transformed = pca.fit_transform(iris.data)
classifier.fit(traindat[:, :k], trainclas)
print('Accuracy neural net sklearn:', round(accuracy, 2), '%.')
cpu_nn_skl[i,iv] = t1 - t0
acc_nn_skl[i,iv] = accuracy
#!!!!!!!!! K_nearest neighbour #!!!!!!!!!!!!!!!
t0 = time.time()
opknn = cpc.k_test(test_features,train_features,train_labels,distance=2,k=3)
t1 = time.time()
predictions = np.array(opknn)
ntot = np.shape(predictions)[0]
ncorrect = np.shape(np.where(predictions == test_labels)[0])[0]
accuracy = 100 * (1.*ncorrect/ntot)
print('Accuracy knn:', round(accuracy, 2), '%.')
cpu_knn[i,iv] = t1 - t0
acc_knn[i,iv] = accuracy
id = np.random.randint(0, nsamp, ntest)
d_test = dat[id, :]
c_test_true = clas[id]
d_train = np.delete(dat, id, 0)
c_train = np.delete(clas, id)
print 'extracting subsample of points for training and classification test'
for i in range(ntest):
print id[i], d_test[i, :]
print ''
print 'fitting k nearest neighbour'
#perform a k nearest neighbour algorithm on the test data to predict which level of heart
#disease will be present in a test patient
c_test = hdp.k_test(d_test, d_train, c_train, distance=2, k=3)
print 'comparing the test with true classifications'
for i in range(ntest):
print 'test class: true class ', c_test[i], c_test_true[i]
idmatch = np.where(c_test == c_test_true)[0]
nmatch = np.shape(idmatch)[0]
print 'accuracy...', 1. * nmatch / ntest
#plot the individual pairs of dimensions in the parameter space to investigate the correlations
if (diagplots == 1):
os.system('rm -rf diagplots')
os.system('mkdir diagplots')
idclas = np.unique(clas)
nclass = np.shape(idclas)[0]