def generate_visualisations(clf, X, y, ts, labels, features):
from bhealth.visualisations import labels_figure
from bhealth.visualisations import features_figure
y_pred = clf.predict(X)
fig = plt.figure(figsize=(10, 4))
ax = fig.add_subplot(2, 1, 1)
fig, ax = labels_figure(y_pred, ts=ts, labels=labels, fig=fig, ax=ax)
ax = fig.add_subplot(2, 1, 2)
df_X = pd.DataFrame(X, index=ts).resample('15Min').agg('mean')
fig, ax = features_figure(df_X.values, ts=df_X.index, feature_names=features, fig=fig, ax=ax)
fig.savefig('predictions.svg')
# Ongoing example of polar plot
from bhealth.visualisations import polar_labels_figure
def most_common(x):
if len(x) == 0:
return -1
return np.argmax(np.bincount(x))
resample = '15Min'
df_labels = pd.DataFrame(y_pred, columns=['label'], index=ts).resample(resample).agg(most_common)
# Add NaN at the beginning for days before installation
first_day = df_labels.index[0].replace(hour=0, minute=0, second=0,
microsecond=0)
previous_monday = first_day + timedelta(days=-first_day.weekday())
leading_labels = (first_day - previous_monday) / resample
df_labels = pd.concat([pd.DataFrame(data=-1, index=[previous_monday],
columns=['label']),
df_labels])
df_labels = df_labels.resample(resample).agg('first')
df_labels = df_labels.fillna(-1)
# Number of columns in one week
n_columns = int(pd.Timedelta('1D') /
pd.Timedelta(resample))
xticklabels = ('24', '1', '2', '3', '4', '5', '6', '7',
'8', '9', '10', '11', '12', '13', '14', '15',
'16', '17', '18', '19', '20', '21', '22', '23')
filename = 'labels_daily.png'
# Add carrying -1 (denoting NaNs)
y_labels = df_labels['label'].values
y_labels = np.concatenate((y_labels, -
np.ones(int(np.ceil(len(y_labels)/n_columns)*n_columns)
- len(y_labels))))
# Create a rectangular matrix with (number_weeks x labels_per_week)
# Int this case every week has n_columns
y_labels = y_labels.reshape((-1, n_columns)).astype(int)
fi, ax = polar_labels_figure(y_labels, labels, xticklabels,
empty_rows=4, leading_labels=0, spiral=True,
title="{} per box".format(resample), m=None)
fig.savefig(filename, dpi=300)
metric_array = [
metrics.duration_sitting, metrics.duration_walking,
metrics.duration_washing, metrics.duration_eating,
metrics.duration_sleeping, metrics.duration_studying,
metrics.number_of_unique_activities
]
metric_container, date_container = metrics.run_metric_array(
metric_array, csv='./output/accelerometer_metrics.csv')
return metric_container, date_container
if __name__ == '__main__':
ts, X, y = get_raw_ts_X_y()
features_figure(X[0:X.size:50], feature_names=['X', 'Y', 'Z'])
X, y = preprocess_X_y(ts, X, y)
(X_train, y_train), (X_test, y_test) = split_train_test(X, y)
clf_grid = get_classifier_grid()
clf_grid.fit(X_train, y_train)
print_summary(clf_grid, X_test, y_test)
metric_container_daily, date_container_daily = activity_metrics(
y, ts, 'daily')
figures_dict = plot_metrics(metric_container_daily, date_container_daily)
for key, fig in figures_dict.items():
fig.savefig(os.path.join('./output/', key))
if __name__ == '__main__':
houses = ['A']
for house_ in houses:
ts, X, y = get_raw_ts_X_y(house_)
X, y = preprocess_X_y(ts, X, y)
(X_train, y_train), (X_test, y_test) = split_train_test(X, y)
clf_grid = get_classifier_grid()
clf_grid.fit(X_train, y_train)
print_summary(clf_grid, X_test, y_test)
metric_container_daily, date_container_daily = localisation_metrics(
y, ts, 'daily')
figures_dict = plot_metrics(
metric_container_daily,
date_container_daily,
labels_=['foyer', 'bedroom', 'living_room', 'bathroom'])
fig, ax = features_figure(X,
feature_names=[
'AP1', 'AP2', 'AP3', 'AP4', 'AP5', 'AP6',
'AP7', 'AP8'
])
figures_dict['features'] = fig
for key, fig in figures_dict.items():
fig.savefig(os.path.join('./output/', key))
metrics.duration_eating,
metrics.duration_sleeping,
metrics.duration_studying,
metrics.number_of_unique_activities]
metric_container, date_container = metrics.run_metric_array(
metric_array, csv='./output/synthetic_metrics.csv')
return metric_container, date_container
if __name__ == '__main__':
ts, X, y, labels, features = get_raw_ts_X_y()
ts, X, y = preprocess_X_y(ts, X, y)
(ts_train, X_train, y_train), (ts_test, X_test, y_test) = split_train_test(ts, X, y)
clf_grid = get_classifier_grid()
clf_grid.fit(X_train, y_train)
print_summary(clf_grid, X_test, y_test)
metric_container_daily, date_container_daily = activity_metrics(y, ts,
'daily')
figures_dict = plot_metrics(metric_container_daily, date_container_daily)
fig, ax = features_figure(X[0:X.size:50], ts[0:ts.size:50],
feature_names=['X', 'Y', 'Z'])
figures_dict['features'] = fig
for key, fig in figures_dict.items():
fig.tight_layout()
fig.savefig(os.path.join('./output/', key))
metrics.duration_in_living_room, metrics.walking_speed,
metrics.room_transfers, metrics.number_of_unique_locations
]
metric_container, date_container = metrics.run_metric_array(
metric_array, csv='./output/synthetic_long_metrics.csv')
return metric_container, date_container
if __name__ == '__main__':
ts, X, y, labels, features = get_raw_ts_X_y()
features_figure(X,
ts,
feature_names=[
'rssi bathroom', 'rssi bedroom 1', 'rssi stairs',
'rssi hall', 'rssi kitchen', 'rssi living room'
])
ts, X, y = preprocess_X_y(ts, X, y)
(ts_train, X_train, y_train), (ts_test, X_test,
y_test) = split_train_test(ts, X, y)
clf_grid = get_classifier_grid()
clf_grid.fit(X_train, y_train)
# =============================================================================
# print_summary(clf_grid, X_test, y_test)
# =============================================================================
# =============================================================================
# generate_visualisations(
# clf_grid.best_estimator_,