|
1
|
|
|
""" |
|
2
|
|
|
Summary: |
|
3
|
|
|
Function fetch_and_preprocess from tutorial_pamap2.py helps to fetch and |
|
4
|
|
|
preproces the data. |
|
5
|
|
|
Example function calls in 'Tutorial mcfly on PAMAP2.ipynb' |
|
6
|
|
|
""" |
|
7
|
1 |
|
import numpy as np |
|
8
|
1 |
|
from numpy import genfromtxt |
|
9
|
1 |
|
import pandas as pd |
|
10
|
1 |
|
import matplotlib.pyplot as plt |
|
11
|
1 |
|
from os import listdir |
|
12
|
1 |
|
import os.path |
|
13
|
1 |
|
import zipfile |
|
14
|
1 |
|
import keras |
|
15
|
1 |
|
from keras.utils.np_utils import to_categorical |
|
16
|
1 |
|
import sys |
|
17
|
1 |
|
import six.moves.urllib as urllib |
|
18
|
|
|
|
|
19
|
|
|
|
|
20
|
1 |
|
def split_activities(labels, X, borders=10 * 100): |
|
|
|
|
|
|
21
|
|
|
""" |
|
22
|
|
|
Splits up the data per activity and exclude activity=0. |
|
23
|
|
|
Also remove borders for each activity. |
|
24
|
|
|
Returns lists with subdatasets |
|
25
|
|
|
|
|
26
|
|
|
Parameters |
|
27
|
|
|
---------- |
|
28
|
|
|
labels : numpy array |
|
29
|
|
|
Activity labels |
|
30
|
|
|
X : numpy array |
|
31
|
|
|
Data points |
|
32
|
|
|
borders : int |
|
33
|
|
|
Nr of timesteps to remove from the borders of an activity |
|
34
|
|
|
|
|
35
|
|
|
Returns |
|
36
|
|
|
------- |
|
37
|
|
|
X_list |
|
38
|
|
|
y_list |
|
39
|
|
|
""" |
|
40
|
1 |
|
tot_len = len(labels) |
|
41
|
1 |
|
startpoints = np.where([1] + [labels[i] != labels[i - 1] |
|
42
|
|
|
for i in range(1, tot_len)])[0] |
|
43
|
1 |
|
endpoints = np.append(startpoints[1:] - 1, tot_len - 1) |
|
44
|
1 |
|
acts = [labels[s] for s, e in zip(startpoints, endpoints)] |
|
45
|
|
|
# Also split up the data, and only keep the non-zero activities |
|
46
|
1 |
|
xysplit = [(X[s + borders:e - borders + 1, :], a) |
|
47
|
|
|
for s, e, a in zip(startpoints, endpoints, acts) if a != 0] |
|
48
|
1 |
|
xysplit = [(X, y) for X, y in xysplit if len(X) > 0] |
|
49
|
1 |
|
Xlist = [X for X, y in xysplit] |
|
|
|
|
|
|
50
|
1 |
|
ylist = [y for X, y in xysplit] |
|
51
|
1 |
|
return Xlist, ylist |
|
52
|
|
|
|
|
53
|
|
|
|
|
54
|
1 |
|
def sliding_window(frame_length, step, Xsamples, |
|
|
|
|
|
|
55
|
|
|
ysamples, Xsampleslist, ysampleslist): |
|
56
|
|
|
""" |
|
57
|
|
|
Splits time series in ysampleslist and Xsampleslist |
|
58
|
|
|
into segments by applying a sliding overlapping window |
|
59
|
|
|
of size equal to frame_length with steps equal to step |
|
60
|
|
|
it does this for all the samples and appends all the output together. |
|
61
|
|
|
So, the participant distinction is not kept |
|
62
|
|
|
|
|
63
|
|
|
Parameters |
|
64
|
|
|
---------- |
|
65
|
|
|
frame_length : int |
|
66
|
|
|
Length of sliding window |
|
67
|
|
|
step : int |
|
68
|
|
|
Stepsize between windows |
|
69
|
|
|
Xsamples : list |
|
70
|
|
|
Existing list of window fragments |
|
71
|
|
|
ysamples : list |
|
72
|
|
|
Existing list of window fragments |
|
73
|
|
|
Xsampleslist : list |
|
74
|
|
|
Samples to take sliding windows from |
|
75
|
|
|
ysampleslist |
|
76
|
|
|
Samples to take sliding windows from |
|
77
|
|
|
|
|
78
|
|
|
""" |
|
79
|
1 |
|
for j in range(len(Xsampleslist)): |
|
80
|
1 |
|
X = Xsampleslist[j] |
|
|
|
|
|
|
81
|
1 |
|
ybinary = ysampleslist[j] |
|
82
|
1 |
|
for i in range(0, X.shape[0] - frame_length, step): |
|
83
|
1 |
|
xsub = X[i:i + frame_length, :] |
|
84
|
1 |
|
ysub = ybinary |
|
85
|
1 |
|
Xsamples.append(xsub) |
|
86
|
1 |
|
ysamples.append(ysub) |
|
87
|
|
|
|
|
88
|
|
|
|
|
89
|
1 |
|
def transform_y(y, mapclasses, nr_classes): |
|
|
|
|
|
|
90
|
|
|
""" |
|
91
|
|
|
Transforms y, a list with one sequence of A timesteps |
|
92
|
|
|
and B unique classes into a binary Numpy matrix of |
|
93
|
|
|
shape (A, B) |
|
94
|
|
|
|
|
95
|
|
|
Parameters |
|
96
|
|
|
---------- |
|
97
|
|
|
y : list or array |
|
98
|
|
|
List of classes |
|
99
|
|
|
mapclasses : dict |
|
100
|
|
|
dictionary that maps the classes to numbers |
|
101
|
|
|
nr_classes : int |
|
102
|
|
|
total number of classes |
|
103
|
|
|
""" |
|
104
|
1 |
|
ymapped = np.array([mapclasses[c] for c in y], dtype='int') |
|
105
|
1 |
|
ybinary = to_categorical(ymapped, nr_classes) |
|
106
|
1 |
|
return ybinary |
|
107
|
|
|
|
|
108
|
|
|
|
|
109
|
1 |
|
def addheader(datasets): |
|
110
|
|
|
""" |
|
111
|
|
|
The columns of the pandas data frame are numbers |
|
112
|
|
|
this function adds the column labels |
|
113
|
|
|
|
|
114
|
|
|
Parameters |
|
115
|
|
|
---------- |
|
116
|
|
|
datasets : list |
|
117
|
|
|
List of pandas dataframes |
|
118
|
|
|
""" |
|
119
|
1 |
|
axes = ['x', 'y', 'z'] |
|
120
|
1 |
|
IMUsensor_columns = ['temperature'] + \ |
|
|
|
|
|
|
121
|
|
|
['acc_16g_' + i for i in axes] + \ |
|
122
|
|
|
['acc_6g_' + i for i in axes] + \ |
|
123
|
|
|
['gyroscope_' + i for i in axes] + \ |
|
124
|
|
|
['magnometer_' + i for i in axes] + \ |
|
125
|
|
|
['orientation_' + str(i) for i in range(4)] |
|
126
|
1 |
|
header = ["timestamp", "activityID", "heartrate"] + ["hand_" + s |
|
127
|
|
|
for s in IMUsensor_columns] \ |
|
|
|
|
|
|
128
|
|
|
+ ["chest_" + s for s in IMUsensor_columns] + ["ankle_" + s |
|
129
|
|
|
for s in IMUsensor_columns] |
|
|
|
|
|
|
130
|
1 |
|
for i in range(0, len(datasets)): |
|
131
|
1 |
|
datasets[i].columns = header |
|
132
|
1 |
|
return datasets |
|
133
|
|
|
|
|
134
|
|
|
|
|
135
|
1 |
|
def numpify_and_store(X, y, xname, yname, outdatapath, shuffle=False): |
|
|
|
|
|
|
136
|
|
|
""" |
|
137
|
|
|
Converts python lists x 3D and y 1D into numpy arrays |
|
138
|
|
|
and stores the numpy array in directory outdatapath |
|
139
|
|
|
shuffle is optional and shuffles the samples |
|
140
|
|
|
|
|
141
|
|
|
Parameters |
|
142
|
|
|
---------- |
|
143
|
|
|
X : list |
|
144
|
|
|
list with data |
|
145
|
|
|
y : list |
|
146
|
|
|
list with data |
|
147
|
|
|
xname : str |
|
148
|
|
|
name to store the x arrays |
|
149
|
|
|
yname : str |
|
150
|
|
|
name to store the y arrays |
|
151
|
|
|
outdatapath : str |
|
152
|
|
|
path to the directory to store the data |
|
153
|
|
|
shuffle : bool |
|
154
|
|
|
whether to shuffle the data before storing |
|
155
|
|
|
""" |
|
156
|
1 |
|
X = np.array(X) |
|
157
|
1 |
|
y = np.array(y) |
|
158
|
|
|
# Shuffle the train set |
|
159
|
1 |
|
if shuffle is True: |
|
160
|
1 |
|
np.random.seed(123) |
|
161
|
1 |
|
neworder = np.random.permutation(X.shape[0]) |
|
162
|
1 |
|
X = X[neworder, :, :] |
|
163
|
1 |
|
y = y[neworder, :] |
|
164
|
|
|
# Save binary file |
|
165
|
1 |
|
np.save(outdatapath + xname, X) |
|
166
|
1 |
|
np.save(outdatapath + yname, y) |
|
167
|
|
|
|
|
168
|
|
|
|
|
169
|
1 |
|
def fetch_data(directory_to_extract_to): |
|
170
|
|
|
""" |
|
171
|
|
|
Fetch the data and extract the contents of the zip file |
|
172
|
|
|
to the directory_to_extract_to. |
|
173
|
|
|
First check whether this was done before, if yes, then skip |
|
174
|
|
|
|
|
175
|
|
|
Parameters |
|
176
|
|
|
---------- |
|
177
|
|
|
directory_to_extract_to : str |
|
178
|
|
|
directory to create subfolder 'PAMAP2' |
|
179
|
|
|
|
|
180
|
|
|
Returns |
|
181
|
|
|
------- |
|
182
|
|
|
targetdir: str |
|
183
|
|
|
directory where the data is extracted |
|
184
|
|
|
""" |
|
185
|
|
|
targetdir = directory_to_extract_to + '/PAMAP2' |
|
186
|
|
|
if os.path.exists(targetdir): |
|
187
|
|
|
print('Data previously downloaded and stored in ' + targetdir) |
|
188
|
|
|
else: |
|
189
|
|
|
os.makedirs(targetdir) # create target directory |
|
190
|
|
|
# Download the PAMAP2 data, this is 688 Mb |
|
191
|
|
|
path_to_zip_file = directory_to_extract_to + '/PAMAP2_Dataset.zip' |
|
192
|
|
|
test_file_exist = os.path.isfile(path_to_zip_file) |
|
193
|
|
|
if test_file_exist is False: |
|
194
|
|
|
url = str('https://archive.ics.uci.edu/ml/' + |
|
195
|
|
|
'machine-learning-databases/00231/PAMAP2_Dataset.zip') |
|
196
|
|
|
# retrieve data from url |
|
197
|
|
|
local_fn, headers = urllib.request.urlretrieve(url, |
|
198
|
|
|
filename=path_to_zip_file) |
|
|
|
|
|
|
199
|
|
|
print('Download complete and stored in: ' + path_to_zip_file) |
|
200
|
|
|
else: |
|
201
|
|
|
print('The data was previously downloaded and stored in ' + |
|
202
|
|
|
path_to_zip_file) |
|
203
|
|
|
# unzip |
|
204
|
|
|
with zipfile.ZipFile(path_to_zip_file, "r") as zip_ref: |
|
205
|
|
|
zip_ref.extractall(targetdir) |
|
206
|
|
|
return targetdir |
|
207
|
|
|
|
|
208
|
|
|
|
|
209
|
1 |
|
def slidingwindow_store(y_list, x_list, X_name, y_name, outdatapath, shuffle): |
|
|
|
|
|
|
210
|
|
|
""" |
|
211
|
|
|
Take sliding-window frames. Target is label of last time step |
|
212
|
|
|
Data is 100 Hz |
|
213
|
|
|
|
|
214
|
|
|
Parameters |
|
215
|
|
|
---------- |
|
216
|
|
|
y_list : list |
|
217
|
|
|
list of arrays with classes |
|
218
|
|
|
x_list : list |
|
219
|
|
|
list of numpy arrays with data |
|
220
|
|
|
X_name : str |
|
221
|
|
|
Name for X file |
|
222
|
|
|
y_name : str |
|
223
|
|
|
Name for y file |
|
224
|
|
|
outdatapath : str |
|
225
|
|
|
directory to store the data |
|
226
|
|
|
shuffle : bool |
|
227
|
|
|
whether to shuffle the data |
|
228
|
|
|
""" |
|
229
|
|
|
frame_length = int(5.12 * 100) |
|
230
|
|
|
step = 1 * 100 |
|
231
|
|
|
x_set = [] |
|
232
|
|
|
y_set = [] |
|
233
|
|
|
sliding_window(frame_length, step, x_set, y_set, x_list, y_list) |
|
234
|
|
|
numpify_and_store(x_set, y_set, X_name, y_name, |
|
235
|
|
|
outdatapath, shuffle) |
|
236
|
|
|
|
|
237
|
|
|
|
|
238
|
1 |
|
def map_class(datasets_filled): |
|
239
|
|
|
ysetall = [set(np.array(data.activityID)) - set([0]) |
|
240
|
|
|
for data in datasets_filled] |
|
241
|
|
|
classlabels = list(set.union(*[set(y) for y in ysetall])) |
|
242
|
|
|
nr_classes = len(classlabels) |
|
243
|
|
|
mapclasses = {classlabels[i]: i for i in range(len(classlabels))} |
|
244
|
|
|
return classlabels, nr_classes, mapclasses |
|
245
|
|
|
|
|
246
|
|
|
|
|
247
|
1 |
|
def split_data(Xlists, ybinarylists, indices): |
|
|
|
|
|
|
248
|
|
|
""" Function takes subset from list given indices |
|
249
|
|
|
|
|
250
|
|
|
Parameters |
|
251
|
|
|
---------- |
|
252
|
|
|
Xlists: tuple |
|
253
|
|
|
tuple (samples) of lists (windows) of numpy-arrays (time, variable) |
|
254
|
|
|
ybinarylist : |
|
255
|
|
|
list (samples) of numpy-arrays (window, class) |
|
256
|
|
|
indices : |
|
257
|
|
|
indices of the slice of data (samples) to be taken |
|
258
|
|
|
|
|
259
|
|
|
Returns |
|
260
|
|
|
------- |
|
261
|
|
|
x_setlist : list |
|
262
|
|
|
list (windows across samples) of numpy-arrays (time, variable) |
|
263
|
|
|
y_setlist: list |
|
264
|
|
|
list (windows across samples) of numpy-arrays (class, ) |
|
265
|
|
|
""" |
|
266
|
1 |
|
tty = str(type(indices)) |
|
267
|
|
|
# or statement in next line is to account for python2 and python3 |
|
268
|
|
|
# difference |
|
269
|
1 |
|
if tty == "<class 'slice'>" or tty == "<type 'slice'>": |
|
270
|
1 |
|
x_setlist = [X for Xlist in Xlists[indices] for X in Xlist] |
|
271
|
1 |
|
y_setlist = [y for ylist in ybinarylists[indices] for y in ylist] |
|
272
|
|
|
else: |
|
273
|
|
|
x_setlist = [X for X in Xlists[indices]] |
|
274
|
|
|
y_setlist = [y for y in ybinarylists[indices]] |
|
275
|
1 |
|
return x_setlist, y_setlist |
|
276
|
|
|
|
|
277
|
|
|
|
|
278
|
1 |
|
def preprocess(targetdir, outdatapath, columns_to_use): |
|
279
|
|
|
""" Function to preprocess the PAMAP2 data after it is fetched |
|
280
|
|
|
|
|
281
|
|
|
Parameters |
|
282
|
|
|
---------- |
|
283
|
|
|
targetdir : str |
|
284
|
|
|
subdirectory of directory_to_extract_to, targetdir |
|
285
|
|
|
is defined by function fetch_data |
|
286
|
|
|
outdatapath : str |
|
287
|
|
|
a subdirectory of directory_to_extract_to, outdatapath |
|
288
|
|
|
is the direcotry where the Numpy output will be stored. |
|
289
|
|
|
columns_to_use : list |
|
290
|
|
|
list of column names to use |
|
291
|
|
|
|
|
292
|
|
|
Returns |
|
293
|
|
|
------- |
|
294
|
|
|
None |
|
295
|
|
|
""" |
|
296
|
|
|
datadir = targetdir + '/PAMAP2_Dataset/Protocol' |
|
297
|
|
|
filenames = listdir(datadir) |
|
298
|
|
|
print('Start pre-processing all ' + str(len(filenames)) + ' files...') |
|
299
|
|
|
# load the files and put them in a list of pandas dataframes: |
|
300
|
|
|
datasets = [pd.read_csv(datadir + '/' + fn, header=None, sep=' ') |
|
301
|
|
|
for fn in filenames] |
|
302
|
|
|
datasets = addheader(datasets) # add headers to the datasets |
|
303
|
|
|
# Interpolate dataset to get same sample rate between channels |
|
304
|
|
|
datasets_filled = [d.interpolate() for d in datasets] |
|
305
|
|
|
# Create mapping for class labels |
|
306
|
|
|
classlabels, nr_classes, mapclasses = map_class(datasets_filled) |
|
307
|
|
|
# Create input (x) and output (y) sets |
|
308
|
|
|
xall = [np.array(data[columns_to_use]) for data in datasets_filled] |
|
309
|
|
|
yall = [np.array(data.activityID) for data in datasets_filled] |
|
310
|
|
|
xylists = [split_activities(y, x) for x, y in zip(xall, yall)] |
|
311
|
|
|
Xlists, ylists = zip(*xylists) |
|
|
|
|
|
|
312
|
|
|
ybinarylists = [transform_y(y, mapclasses, nr_classes) for y in ylists] |
|
313
|
|
|
# Split in train, test and val |
|
314
|
|
|
x_vallist, y_vallist = split_data(Xlists, ybinarylists, indices=6) |
|
315
|
|
|
test_range = slice(7, len(datasets_filled)) |
|
316
|
|
|
x_testlist, y_testlist = split_data(Xlists, ybinarylists, test_range) |
|
317
|
|
|
x_trainlist, y_trainlist = split_data(Xlists, ybinarylists, |
|
318
|
|
|
indices=slice(0, 6)) |
|
319
|
|
|
# Take sliding-window frames, target is label of last time step, |
|
320
|
|
|
# and store as numpy file |
|
321
|
|
|
slidingwindow_store(y_list=y_trainlist, x_list=x_trainlist, |
|
322
|
|
|
X_name='X_train', y_name='y_train', |
|
323
|
|
|
outdatapath=outdatapath, shuffle=True) |
|
324
|
|
|
slidingwindow_store(y_list=y_vallist, x_list=x_vallist, |
|
325
|
|
|
X_name='X_val', y_name='y_val', |
|
326
|
|
|
outdatapath=outdatapath, shuffle=False) |
|
327
|
|
|
slidingwindow_store(y_list=y_testlist, x_list=x_testlist, |
|
328
|
|
|
X_name='X_test', y_name='y_test', |
|
329
|
|
|
outdatapath=outdatapath, shuffle=False) |
|
330
|
|
|
print('Processed data succesfully stored in ' + outdatapath) |
|
331
|
|
|
return None |
|
332
|
|
|
|
|
333
|
|
|
|
|
334
|
1 |
|
def fetch_and_preprocess(directory_to_extract_to, columns_to_use=None): |
|
335
|
|
|
""" |
|
336
|
|
|
High level function to fetch_and_preprocess the PAMAP2 dataset |
|
337
|
|
|
|
|
338
|
|
|
Parameters |
|
339
|
|
|
---------- |
|
340
|
|
|
directory_to_extract_to : str |
|
341
|
|
|
the directory where the data will be stored |
|
342
|
|
|
columns_to_use : list |
|
343
|
|
|
the columns to use |
|
344
|
|
|
|
|
345
|
|
|
Returns |
|
346
|
|
|
------- |
|
347
|
|
|
outdatapath: str |
|
348
|
|
|
The directory in which the numpy files are stored |
|
349
|
|
|
""" |
|
350
|
|
|
if columns_to_use is None: |
|
351
|
|
|
columns_to_use = ['hand_acc_16g_x', 'hand_acc_16g_y', 'hand_acc_16g_z', |
|
352
|
|
|
'ankle_acc_16g_x', 'ankle_acc_16g_y', 'ankle_acc_16g_z', |
|
|
|
|
|
|
353
|
|
|
'chest_acc_16g_x', 'chest_acc_16g_y', 'chest_acc_16g_z'] |
|
|
|
|
|
|
354
|
|
|
targetdir = fetch_data(directory_to_extract_to) |
|
355
|
|
|
outdatapath = targetdir + '/PAMAP2_Dataset/slidingwindow512cleaned/' |
|
356
|
|
|
if not os.path.exists(outdatapath): |
|
357
|
|
|
os.makedirs(outdatapath) |
|
358
|
|
|
if os.path.isfile(outdatapath + 'x_train.npy'): |
|
359
|
|
|
print('Data previously pre-processed and np-files saved to ' + |
|
360
|
|
|
outdatapath) |
|
361
|
|
|
else: |
|
362
|
|
|
preprocess(targetdir, outdatapath, columns_to_use) |
|
363
|
|
|
return outdatapath |
|
364
|
|
|
|
|
365
|
|
|
|
|
366
|
1 |
|
def load_data(outputpath): |
|
367
|
|
|
""" Function to load the numpy data as stored in directory |
|
368
|
|
|
outputpath. |
|
369
|
|
|
|
|
370
|
|
|
Parameters |
|
371
|
|
|
---------- |
|
372
|
|
|
outputpath : str |
|
373
|
|
|
directory where the numpy files are stored |
|
374
|
|
|
|
|
375
|
|
|
Returns |
|
376
|
|
|
------- |
|
377
|
|
|
x_train |
|
378
|
|
|
y_train_binary |
|
379
|
|
|
x_val |
|
380
|
|
|
y_val_binary |
|
381
|
|
|
x_test |
|
382
|
|
|
y_test_binary |
|
383
|
|
|
""" |
|
384
|
|
|
ext = '.npy' |
|
385
|
|
|
x_train = np.load(outputpath + 'X_train' + ext) |
|
386
|
|
|
y_train_binary = np.load(outputpath + 'y_train' + ext) |
|
387
|
|
|
x_val = np.load(outputpath + 'X_val' + ext) |
|
388
|
|
|
y_val_binary = np.load(outputpath + 'y_val' + ext) |
|
389
|
|
|
x_test = np.load(outputpath + 'X_test' + ext) |
|
390
|
|
|
y_test_binary = np.load(outputpath + 'y_test' + ext) |
|
391
|
|
|
return x_train, y_train_binary, x_val, y_val_binary, x_test, y_test_binary |
|
392
|
|
|
|
NLeSC /
mcfly
Loading history...
This check looks for invalid names for a range of different identifiers.
You can set regular expressions to which the identifiers must conform if the defaults do not match your requirements.
If your project includes a Pylint configuration file, the settings contained in that file take precedence.
To find out more about Pylint, please refer to their site.