def test_ae_class(self, objData, normalize=False):
if normalize is True:
objData.normalization(self.normalization_max)
minibatch_aux = objData.minibatch
objData.change_minibatch(1)
y_true = objData.labels
y_result = []
print('\n# TEST AENCODER BY CLASS TRAINED')
for ix in range(objData.total_batchs_complete):
x_, label = objData.generate_batch()
cost_class = []
for class_i in range(self.num_class):
cost_i = self.sess.run(self.AEclass[class_i].cost,
feed_dict={self.x_batch: x_})
cost_class.append(cost_i)
y_result.append(np.argsort(cost_class)[0])
objData.next_batch_test()
objData.change_minibatch(minibatch_aux)
utils.metrics_multiclass(y_true, y_result)
def test_vgg(self, objData, normalize=False):
if normalize is True:
objData.normalization(self.normalization_max)
count_success = 0
prob_predicted = []
plot_predicted = []
label_total = []
prob_total = np.random.random((0, self.num_class))
print('\n# TEST VGG TRAINED')
for i in range(objData.total_batchs_complete):
batch, label = objData.generate_batch()
prob = self.sess.run(self.probVGG, feed_dict={self.x_batch: batch})
label_total = np.concatenate((label_total, label), axis=0)
prob_total = np.concatenate((prob_total, prob), axis=0)
# Acumulamos la presicion de cada iteracion, para despues hacer un promedio
count, prob_predicted, plot_predicted = utils.process_prob(
label,
prob,
predicted=prob_predicted,
plot_predicted=plot_predicted)
count_success = count_success + count
objData.next_batch_test()
# promediamos la presicion total
print('\n# STATUS:')
y_true = objData.labels
y_prob = prob_predicted
utils.metrics_multiclass(y_true, y_prob)
def test_model(net, session, objData, writer=None):
label_total = []
label_total_pred = []
print('\n # PHASE: Test classification')
for i in range(objData.total_batchs_complete):
batch, label = objData.generate_batch()
# target = tf.one_hot(label, on_value=1, off_value=0, depth=net.num_class)
# target = list(session.run(target))
label_pred, acc, count, fc6 = session.run([net.labels_pred, net.accuracy, net.correct_count, net.fc6], feed_dict={vgg_batch: batch, vgg_label: label, train_mode: False})
label_total = np.concatenate((label_total, label), axis=0)
label_total_pred = np.concatenate((label_total_pred, label_pred), axis=0)
total_batch = len(label)
objData.next_batch_test()
net.add_row_data_by_save(fc6, label, i)
# writer.add_summary(summary, i)
print(' results[ Total:'+str(total_batch)+' | True:'+str(count)+' | False:'+str(total_batch-count)+' | Accuracy:'+str(acc)+' ]')
# Promediamos la presicion total
print('\n # STATUS:')
y_true = label_total
y_prob = label_total_pred
accuracy_final = metrics_multiclass(y_true, y_prob)
return accuracy_final
def test_model(net, sess_test, objData):
total = objData.total_images
count_success = 0
count_by_class = np.zeros([net.num_class, net.num_class])
prob_predicted = []
plot_predicted = []
label_total = []
prob_total = np.random.random((0, net.num_class))
print('\n# PHASE: Test classification')
for i in range(objData.total_batchs_complete):
batch, label = objData.generate_batch()
prob, layer = sess_test.run([net.prob, net.relu6], feed_dict={vgg_batch: batch, train_mode: False})
label_total = np.concatenate((label_total, label), axis=0)
prob_total = np.concatenate((prob_total, prob), axis=0)
# save output of a layer
# utils.save_layer_output(layer, label, name='Train_SNC4_relu6', dir='../data/features/')
# utils.save_layer_output_by_class(layer, label, name='Train_SNC4', dir='../data/features/')
count, prob_predicted, plot_predicted = utils.process_prob(label, prob, predicted=prob_predicted,
plot_predicted=plot_predicted)
count_success = count_success + count
objData.next_batch_test()
# promediamos la presicion total
print('\n# STATUS:')
y_true = objData.labels
y_prob = prob_predicted
accuracy_final = utils.metrics_multiclass(y_true, y_prob)
return accuracy_final
def test_model_all(net, sess_test, objData, numClass):
y_true = objData.labels
y_result = []
for ix in range(objData.total_batchs_complete):
x_, label = objData.generate_batch()
cost_class = []
for class_i in range(numClass):
cost_i = sess_test.run(net[class_i].cost, feed_dict={x_batch: x_})
cost_class.append(cost_i)
y_result.append(np.argsort(cost_class)[0])
objData.next_batch_test()
utils.metrics_multiclass(y_true, y_result)
def test_model_all(net, sess_test, objData, numClass):
y_true = objData.labels
y_result = []
for ix in range(objData.total_batchs_complete):
x_, label = objData.generate_batch()
cost_class = []
for class_i in range(numClass):
cost_i = sess_test.run(net[class_i].cost, feed_dict={x_batch: x_})
cost_class.append(cost_i)
y_result.append(np.argsort(cost_class)[0])
objData.next_batch_test()
# print(np.shape(y_true))
# ax = np.reshape(objData.labels.values, [objData.total_inputs])
# ax = list(ax)
# f = open("PaperCIARP.csv", "a+")
# f.write(",".join(map(str, ax)) + "\n")
# f.write(",".join(map(str, y_result)) + "\n")
# f.close()
utils.metrics_multiclass(y_true, y_result)
for i in range(0, 3):
path_data_train_csv, path_data_test_csv, path_max_csv, name = path_datasets(
i)
print('\n[NAME:', name, ']')
Damax = utils.load_max_csvData(path_max_csv)
# Metodo 1
# X_train, X_test, y_train, y_test, total_train, total_test = get_data_split(path_data_test_csv, Damax, 0.3)
# Metodo 2
X_train, X_test, y_train, y_test, total_train, total_test = get_data_all(
path_data_train_csv, path_data_test_csv, Damax)
print(np.shape(X_train), np.shape(X_test))
knn = neighbors.KNeighborsClassifier()
print(" Train model...")
knn.fit(X_train, y_train)
print(" Test model...")
Z = knn.predict(X_test)
acc = utils.metrics_multiclass(y_test, Z)
print(' Save result...')
output = [name, total_test, acc, path_data_test_csv]
f.write(','.join(map(str, output)) + '\n')
f.write(','.join(map(str, y_test)) + '\n')
f.write(','.join(map(str, Z)) + '\n')
f.close()
