print('\nPrecision score: %f \n' % precision)
print('\nF-measure score: %f \n' % f_measure)
print('\nRecall score: %f \n' % recall)
print('Window size 1: %f \n' % max_win1)
print('Window size 2: %f \n' % max_win2)
print('Window size 3: %f \n' % max_win3)
print('Window size 4: %f \n' % max_win4)
print('Window size 5: %f \n' % max_win5)
print('Window size 6: %f \n' % max_win6)
print('Window size filter: %f \n' % max_win_filter)
max_win = max(max_win1, max_win2, max_win3, max_win4, max_win5, max_win6)
# Prune the original sensor data.
for i in range(0,cv_x_rep.shape[0]):
if i<=((max_win + max_win_filter - 2)/2 - 1) or i>=(cv_x_rep.shape[0]-(max_win + max_win_filter - 2)/2):
cv_x_piezo = cv_x_piezo.drop([i])
cv_x_strain = cv_x_strain.drop([i])
cv_x_mic = cv_x_mic.drop([i])
cv_x_piezo.index = range(0,cv_x_piezo.shape[0])
cv_x_strain.index = range(0,cv_x_strain.shape[0])
cv_x_mic.index = range(0,cv_x_mic.shape[0])
# Visualize the results.
plot.draw_coloredLabel(cv_y_mark, predicted_mark, (-500, cv_y_mark.shape[0]+5000))
# plot.draw_label(cv_y_mark, predicted_mark, (-500, cv_y_mark.shape[0]+500))
plot.draw_sensorData(cv_x_piezo, cv_y_mark, predicted_mark, (-500, cv_x_piezo.shape[0]+4000))
plot.draw_sensorData(cv_x_strain, cv_y_mark, predicted_mark, (-500, cv_x_piezo.shape[0]+4000))
plot.draw_sensorData(cv_x_mic, cv_y_mark, predicted_mark, (-500, cv_x_piezo.shape[0]+4000))
def main(argv):
args = parser.parse_args(argv[1:])
# Fetch the data.
(cv_x, cv_y, cv_x_df) = sensor_data.load_data()
cv_y.index = range(0, cv_y.shape[0])
# Feature columns describe how to use the input.
my_feature_columns = []
for key in cv_x_df.keys():
my_feature_columns.append(tf.feature_column.numeric_column(key=key))
# Build 2 hidden layer DNN with 10, 10 units respectively.
classifier = tf.estimator.DNNClassifier(feature_columns=my_feature_columns,
hidden_units=[10, 10],
n_classes=5)
# Cross validation
n_splits = 50
dataset = sensor_data.cv_dataset(cv_x, cv_y, n_splits)
accu = 0
step = 0
pred_label = []
print(type(cv_y))
for x_train, y_train, x_test, y_test in dataset:
x_train = pd.DataFrame(x_train,
columns=[
'Mean_Piezo', 'Std_Piezo', 'Mean_Strain',
'Std_Strain', 'Mean_Mic', 'Std_Mic'
])
y_train = pd.DataFrame(y_train)
x_test = pd.DataFrame(x_test,
columns=[
'Mean_Piezo', 'Std_Piezo', 'Mean_Strain',
'Std_Strain', 'Mean_Mic', 'Std_Mic'
])
y_test = pd.DataFrame(y_test)
step = step + 1
classifier.train(input_fn=lambda: sensor_data.train_input_fn(
x_train, y_train, args.batch_size),
steps=args.train_steps)
predictions = classifier.predict(
input_fn=lambda: sensor_data.eval_input_fn(
x_test, labels=None, batch_size=args.batch_size))
for pred_dict in predictions:
pred_label.append(pred_dict['class_ids'][0])
pred_label = pd.Series(pred_label)
# Apply mode filter to the estimated labels.
filter_mode = lambda x: mode(x)[0][0]
window_filter = 51
pred_label = pred_label.rolling(window=window_filter,
center=True).apply(filter_mode, raw=True)
cv_y.index = range(0, cv_y.shape[0])
pred_label.index = range(0, pred_label.shape[0])
# Eliminate the begining and the end of the labels after the mode filtering.
for i in range(0, cv_y.shape[0]):
if i <= (
(window_filter - 1) / 2 - 1) or i >= (cv_y.shape[0] -
(window_filter - 1) / 2):
cv_y = cv_y.drop([i])
pred_label = pred_label.drop([i])
cv_y.index = range(0, cv_y.shape[0])
pred_label.index = range(0, pred_label.shape[0])
precision = metrics.precision_score(cv_y, pred_label, average='weighted')
f_measure = metrics.f1_score(cv_y, pred_label, average='weighted')
recall = metrics.recall_score(cv_y, pred_label, average='weighted')
# Calculate and output the accuracy.
print('\nTest set accuracy: %f \n' % (1 - (((cv_y != pred_label).sum()) /
(cv_y.shape[0]))))
# Output the precision, recall and f-measure scores.f-measure.
print('\nPrecision score: %f \n' % precision)
print('\nRecall score: %f \n' % recall)
print('\nF-measure score: %f \n' % f_measure)
cv_x = sensor_data.sensorData()
cv_x_piezo = cv_x[0]
cv_x_strain = cv_x[1]
cv_x_mic = cv_x[2]
max_window_size = sensor_data.transferWindowSize()
# Prune the original sensor data.
for i in range(0, cv_x_piezo.shape[0]):
if i <= ((max_window_size + window_filter - 2) / 2 -
1) or i >= (cv_x_piezo.shape[0] -
(max_window_size + window_filter - 2) / 2):
cv_x_piezo = cv_x_piezo.drop([i])
cv_x_strain = cv_x_strain.drop([i])
cv_x_mic = cv_x_mic.drop([i])
cv_x_piezo.index = range(0, cv_x_piezo.shape[0])
cv_x_strain.index = range(0, cv_x_strain.shape[0])
cv_x_mic.index = range(0, cv_x_mic.shape[0])
# Plot the results.
plot.draw_coloredLabel(cv_y, pred_label, (-500, cv_y.shape[0] + 5000))
plot.draw_sensorData(cv_x_piezo, cv_y, pred_label,
(-500, cv_y.shape[0] + 4000))
plot.draw_sensorData(cv_x_strain, cv_y, pred_label,
(-500, cv_y.shape[0] + 4000))
plot.draw_sensorData(cv_x_mic, cv_y, pred_label,
(-500, cv_y.shape[0] + 4000))