def plot_confusion(classifier, test_pts, test_labels):
classes = [
'STANDING', 'SITTING', 'LYING', 'WALKING', 'WALKING_DOWNSTAIRS',
'WALKING_UPSTAIRS'
]
cl = ['STANDING', 'SITTING', 'LYING', 'WALK', 'WALK_DOWN', 'WALK_UP']
pred_label = classifier.predict(test_pts)
# print(true_label)
result = cf(test_labels, pred_label, labels=classes)
# res_nor = np.ndarray((6, 6), dtype=float)
# for i in range(0, 6):
# s = result[i].sum()
# for j in range(0, 6):
# res_nor[i][j] = float(result[i][j] / s)
print(result)
# print(res_nor)
fig = plt.figure()
ax = fig.add_subplot(111)
cax = ax.matshow(result)
# plt.matshow(result)
fig.colorbar(cax)
ax.set_xticklabels([''] + cl)
ax.set_yticklabels([''] + cl)
plt.xlabel("Predicted Label")
plt.ylabel("True Label")
plt.legend(loc='best')
plt.show()
def plot_confusion(classifier, test_pts, test_labels):
classes = ['STANDING',
'SITTING',
'LYING',
'WALKING',
'WALKING_DOWNSTAIRS',
'WALKING_UPSTAIRS']
cl = ['STANDING',
'SITTING',
'LYING',
'WALK',
'WALK_DOWN',
'WALK_UP']
pred_label = classifier.predict(test_pts)
# print(true_label)
result = cf(test_labels, pred_label, labels=classes)
res_nor = np.ndarray((6, 6), dtype=float)
# for i in range(0, 6):
# s = result[i].sum()
# for j in range(0, 6):
# res_nor[i][j] = float(result[i][j] / s)
print(result)
# print(res_nor)
fig = plt.figure()
ax = fig.add_subplot(111)
cax = ax.matshow(result)
# plt.matshow(result)
fig.colorbar(cax)
ax.set_xticklabels([''] + cl)
ax.set_yticklabels([''] + cl)
plt.xlabel("Predicted Label")
plt.ylabel("True Label")
plt.legend(loc='best')
plt.show()
def print_out(name,one,two): plt.figure() df = pd.DataFrame( cf(one,two), index=labels,columns=labels) sns.heatmap(df,annot=True,cmap='Blues', fmt='g') plt.savefig(name+'.png')
def DTtest(trainData, trainLabel, size):
train_data, test_data, train_label, test_label = \
train_test_split(trainData, trainLabel, test_size=size)
dt = DecisionTreeClassifier()
dt.fit(train_data, train_label)
predict = dt.predict(test_data, test_label)
# number of mix-matched label
error = 0
for x in range(len(predict)):
if predict[x] != test_label[x]:
error += 1
# confusion matrix
cf(test_label, predict)
# accuracy
print(1 - (error / len(predict)))
return error
def classification_report(model, testloader, device=None):
if device is not None:
model.to(device)
y_true = []
y_pred = []
for (images, labels) in testloader:
if device is not None:
images = images.to(device)
labels = labels.to(device)
y_true.append(labels)
y_pred.append(model(images))
y_true = torch.cat(y_true)
y_pred = torch.cat(y_pred)
y_pred = y_pred.argmax(dim=1)
return cf(y_true.cpu().numpy(), y_pred.cpu().numpy())
def classification_report_pgd(model, testloader, device=None, pgd_params={}):
if device is not None:
model.to(device)
y_true = []
y_pred = []
for (images, labels) in testloader:
if device is not None:
images = images.to(device)
labels = labels.to(device)
attacked_images = pgd(model,
nn.CrossEntropyLoss(),
images,
labels,
device=device,
**pgd_params)
y_true.append(labels)
y_pred.append(model(attacked_images))
y_true = torch.cat(y_true)
y_pred = torch.cat(y_pred)
y_pred = y_pred.argmax(dim=1)
return cf(y_true.cpu().numpy(), y_pred.cpu().numpy())
#predicted values from x for possible values of y
Ynewpred = lr.predict(X)
bank[
"Y Predicted"] = Ynewpred # making a new column with these values in bank dataframe
YProbability = pd.DataFrame(
lr.predict_proba(X.iloc[:, :])
) #making a data frame from predicted values of y from using x variables
banknew = pd.concat(
[bank, YProbability],
axis=1) #making new data frame with these y probability values
#forming confusion matrix
from sklearn.metrics import confusion_matrix as cf
cf1 = cf(Y, Ynewpred) #between y original and y predicted by the model
print(cf1)
# y n
#y [39162 760]
#n [ 4226 1063]]
# accuracy = 39162+1063/39612+760+4226+1063 == 0.88
#(39162+1063)/(39612+760+4226+1063)
from sklearn.metrics import roc_curve as rocc
from sklearn.metrics import roc_auc_score as rocarea
fpr, tpr, thresholdasd = rocc(Y, Ynewpred)
auc = rocarea(Y, Ynewpred)
import matplotlib.pyplot as plt
plt.plot(fpr, tpr, color='red', label='ROC')
test_indices, y_test = filter_indices(test_i, labels_reduced)
cv.append((train_indices, labels, test_indices, y_test))
y = [] # initialization of result list
confusion = np.zeros((5, 5)) # confusion matrix initialization
for train_indices, labels, test_indices, y_test in cv:
Rferns = FernEnsemble(ps, fn, fs, dist=dist)
Rferns.train(polsar_reduced, train_indices, labels)
y_pred = Rferns.predict(polsar_reduced, test_indices, prediction='maximum')
y.append(
(y_pred, y_test)
) # for each cv iteration save prediction y_pred and the true labels y_test
confusion += cf(
y_test,
y_pred) # for each iteration add the results to the confusion matrix
name = str(dist) + '_ps' + str(ps) + '_fn' + str(fn) + '_fs' + str(fs)
pickle.dump(y,
open(name + '.p',
"wb")) # save the results as pickle file for later inspection
################################################################################################
####################### Visualization of Output ################################################
################################################################################################
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import numpy as np
from matplotlib import gridspec
#Dropout layer for Convergence
X = Dropout(0.5)(X)
#output a sigmoid to squash the matrix into output probabilities
X = Dense(1, activation='sigmoid')(X)
model = Model(inputs=X_input, outputs=X, name="CNN")
return model
model = model_nn(input_shape=(64, 64, 1))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
# Calling load_dataset to store the dataset
X_train_orig, X_test_orig, Y_train, Y_test = load_dataset()
# Normalize image vectors
X_train = X_train_orig / 255.
X_test = X_test_orig / 255.
model.fit(X_train, Y_train, epochs=15, batch_size=32)
y_pred = model.predict(X_test)
# The output of the model is array of real numbers therefore values greater than 0.5 will be evaluated as 1 otherwise 0
y_pred = (y_pred > 0.5)
# Confusion Matrix
cf(Y_test, y_pred)
# Save the model for further use
model.save('models/CNN_Model.h5', overwrite=True)
def confusion_matrix(self):
assert self.y_test.shape == self.final_pred.shape
matrix = cf(self.y_test, self.final_pred)
print(matrix)
