|
1
|
|
|
import os |
|
2
|
|
|
import pickle |
|
3
|
|
|
import logging |
|
4
|
|
|
import numpy as np |
|
5
|
|
|
from fuel.datasets import H5PYDataset |
|
6
|
|
|
|
|
7
|
|
|
logger = logging.getLogger(__name__) |
|
8
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
class CASASFuel(object): |
|
11
|
|
|
"""CASASFuel Class to retrieve CASAS smart home data as a fuel dataset object |
|
12
|
|
|
|
|
13
|
|
|
Args: |
|
14
|
|
|
dir_name (:obj:`string`): |
|
15
|
|
|
Directory path that contains HDF5 dataset file and complementary dataset information pkl file |
|
16
|
|
|
|
|
17
|
|
|
Attributes: |
|
18
|
|
|
data_filename (:obj:`str`): Path to `data.hdf5` dataset file |
|
19
|
|
|
info (:obj:`dict`): complementary dataset information stored in dict format |
|
20
|
|
|
keys of info includes: |
|
21
|
|
|
""" |
|
22
|
|
|
def __init__(self, dir_name): |
|
23
|
|
|
logger.debug('Load Casas H5PYDataset from ' + dir_name) |
|
24
|
|
|
self.data_filename = dir_name + '/data.hdf5' |
|
25
|
|
|
if os.path.isfile(dir_name + '/info.pkl'): |
|
26
|
|
|
f = open(dir_name + '/info.pkl', 'rb') |
|
27
|
|
|
self.info = pickle.load(f) |
|
28
|
|
|
f.close() |
|
29
|
|
|
else: |
|
30
|
|
|
logger.error('Cannot find info.pkl from current H5PYDataset directory %s' % dir_name) |
|
31
|
|
|
|
|
32
|
|
|
def get_dataset(self, which_sets, load_in_memory=False, **kwargs): |
|
33
|
|
|
"""Return fuel dataset object specified by which_sets tuple and load it in memory |
|
34
|
|
|
|
|
35
|
|
|
Args: |
|
36
|
|
|
which_sets (:obj:`tuple` of :obj:`str`): containing the name of splits to load. |
|
37
|
|
|
Valid value are determined by the ``info.pkl`` loaded. |
|
38
|
|
|
You can get the list of split set names by :meth:`get_set_list()`. |
|
39
|
|
|
Usually, if the dataset is split by weeks, the split name is in the form of ``week <num>``. |
|
40
|
|
|
If the dataset is split by days, the split name is in the form of ``day <num>``. |
|
41
|
|
|
load_in_memory (:obj:`bool`, Optional): Default to False. |
|
42
|
|
|
Whether to load the data in main memory. |
|
43
|
|
|
|
|
44
|
|
|
Returns: |
|
45
|
|
|
:class:`fuel.datasets.base.Dataset`: A Fuel dataset object created by |
|
46
|
|
|
:class:`fuel.datasets.h5py.H5PYDataset` |
|
47
|
|
|
""" |
|
48
|
|
|
# Check if sets exist as split name in metadata |
|
49
|
|
|
for set_name in which_sets: |
|
50
|
|
|
if set_name not in self.info['split_sets']: |
|
51
|
|
|
logger.error('set %s not found in splits' % set_name) |
|
52
|
|
|
# Load specified splits and return |
|
53
|
|
|
return H5PYDataset(file_or_path=self.data_filename, |
|
54
|
|
|
which_sets=which_sets, |
|
55
|
|
|
load_in_memory=load_in_memory, **kwargs) |
|
56
|
|
|
|
|
57
|
|
|
def get_set_list(self): |
|
58
|
|
|
"""Get the split set list |
|
59
|
|
|
|
|
60
|
|
|
Returns: |
|
61
|
|
|
:obj:`tuple` of :obj:`str`: A list of split set names |
|
62
|
|
|
""" |
|
63
|
|
|
return self.info['split_sets'] |
|
64
|
|
|
|
|
65
|
|
|
def get_input_dims(self): |
|
66
|
|
|
"""Get the dimension of features |
|
67
|
|
|
|
|
68
|
|
|
Returns: |
|
69
|
|
|
:obj:`int` : the input feature length |
|
70
|
|
|
""" |
|
71
|
|
|
dims = len(self.info['index_to_feature']) |
|
72
|
|
|
return dims |
|
73
|
|
|
|
|
74
|
|
|
def get_output_dims(self): |
|
75
|
|
|
"""Get the dimension of target indices |
|
76
|
|
|
|
|
77
|
|
|
Returns: |
|
78
|
|
|
:obj:`int` : the target indices |
|
79
|
|
|
""" |
|
80
|
|
|
dims = len(self.info['index_to_activity']) |
|
81
|
|
|
return dims |
|
82
|
|
|
|
|
83
|
|
|
def get_activity_by_index(self, index): |
|
84
|
|
|
"""Get activity name by index |
|
85
|
|
|
|
|
86
|
|
|
Args: |
|
87
|
|
|
index (:obj:`int`): Activity index |
|
88
|
|
|
|
|
89
|
|
|
Returns: |
|
90
|
|
|
:obj:`str`: Activity label |
|
91
|
|
|
""" |
|
92
|
|
|
activity_len = len(self.info['index_to_activity']) |
|
93
|
|
|
if index < activity_len: |
|
94
|
|
|
return self.info['index_to_activity'][index] |
|
95
|
|
|
else: |
|
96
|
|
|
logger.error('Activity index %d out of bound. Dataset has %d activities' % (index, activity_len)) |
|
97
|
|
|
return '' |
|
98
|
|
|
|
|
99
|
|
|
def get_feature_by_index(self, index): |
|
100
|
|
|
"""Get feature string by index |
|
101
|
|
|
|
|
102
|
|
|
Args: |
|
103
|
|
|
index (:obj:`int`): Feature index |
|
104
|
|
|
|
|
105
|
|
|
Returns: |
|
106
|
|
|
:obj:`str`: Feature string |
|
107
|
|
|
""" |
|
108
|
|
|
feature_len = len(self.info['index_to_feature']) |
|
109
|
|
|
if index < feature_len: |
|
110
|
|
|
return self.info['index_to_feature'][index] |
|
111
|
|
|
else: |
|
112
|
|
|
logger.error('Feature index %d out of bound. Dataset has %d features' % (index, feature_len)) |
|
113
|
|
|
return '' |
|
114
|
|
|
|
|
115
|
|
|
def back_annotate(self, fp, prediction, split_id=-1, split_name=None): |
|
116
|
|
|
"""Back annotated predictions of a split set into file pointer |
|
117
|
|
|
|
|
118
|
|
|
Args: |
|
119
|
|
|
fp (:obj:`file`): File object to the back annotation file. |
|
120
|
|
|
prediction (:obj:`numpy.ndarray`): Numpy array containing prediction labels. |
|
121
|
|
|
split_id (:obj:`int`): The index of split set to be annotated (required if split_name not specified). |
|
122
|
|
|
split_name (:obj:`str`): The name of the split set to be annotated (required if split_id is not specified). |
|
123
|
|
|
""" |
|
124
|
|
|
time_array = self._get_time_array(split_id=split_id, split_name=split_name) |
|
125
|
|
|
# Check length of prediction and time array |
|
126
|
|
|
if prediction.shape[0] != len(time_array): |
|
127
|
|
|
logger.error('Prediction size miss-match. There are %d time points with only %d labels given.' % |
|
128
|
|
|
(len(time_array), prediction.shape[0])) |
|
129
|
|
|
return |
|
130
|
|
|
# Perform back annotation |
|
131
|
|
|
for i in range(len(time_array)): |
|
132
|
|
|
if prediction[i] != -1: |
|
133
|
|
|
fp.write('%s %s\n' % (time_array[i].strftime('%Y-%m-%d %H:%M:%S'), |
|
134
|
|
|
self.get_activity_by_index(prediction[i]))) |
|
135
|
|
|
|
|
136
|
|
|
def back_annotate_with_proba(self, fp, prediction_proba, split_id=-1, split_name=None, top_n=-1): |
|
137
|
|
|
"""Back annotated prediction probabilities of a split set into file pointer |
|
138
|
|
|
|
|
139
|
|
|
Args: |
|
140
|
|
|
fp (:obj:`file`): File object to the back annotation file. |
|
141
|
|
|
prediction_proba (:obj:`numpy.ndarray`): Numpy array containing probability for each class in shape |
|
142
|
|
|
of (num_samples, num_class). |
|
143
|
|
|
split_id (:obj:`int`): The index of split set to be annotated (required if split_name not specified). |
|
144
|
|
|
split_name (:obj:`str`): The name of the split set to be annotated (required if split_id is not specified). |
|
145
|
|
|
top_n (:obj:`int`): Back annotate top n probabilities. |
|
146
|
|
|
""" |
|
147
|
|
|
time_array = self._get_time_array(split_id=split_id, split_name=split_name) |
|
148
|
|
|
# Check length of prediction and time array |
|
149
|
|
|
if prediction_proba.shape[0] != len(time_array): |
|
150
|
|
|
logger.error('Prediction size miss-match. There are %d time points with only %d labels given.' % |
|
151
|
|
|
(len(time_array), prediction_proba.shape[0])) |
|
152
|
|
|
return |
|
153
|
|
|
if top_n == -1: |
|
154
|
|
|
top_n = self.get_output_dims() |
|
155
|
|
|
# Perform back annotation |
|
156
|
|
|
for i in range(len(time_array)): |
|
157
|
|
|
sorted_index = np.argsort(prediction_proba[i, :])[::-1] |
|
158
|
|
|
if prediction_proba[i, sorted_index[0]] != -1: |
|
159
|
|
|
fp.write('%s' % time_array[i].strftime('%Y-%m-%d %H:%M:%S')) |
|
160
|
|
|
for j in range(top_n): |
|
161
|
|
|
fp.write(', %s(%g)' % (self.get_activity_by_index(sorted_index[j]), |
|
162
|
|
|
prediction_proba[i, sorted_index[j]])) |
|
163
|
|
|
fp.write('\n') |
|
164
|
|
|
|
|
165
|
|
|
def _get_time_array(self, split_id=-1, split_name=None): |
|
166
|
|
|
"""Get Time Array based for specified splits |
|
167
|
|
|
|
|
168
|
|
|
Args: |
|
169
|
|
|
split_id (:obj:`int`): The index of split set to be annotated (required if split_name not specified). |
|
170
|
|
|
split_name (:obj:`str`): The name of the split set to be annotated (required if split_id is not specified). |
|
171
|
|
|
|
|
172
|
|
|
Returns: |
|
173
|
|
|
:obj:`list` of :obj:`datetime.datetime`: List of event.rst datetime objects of splits specified. |
|
174
|
|
|
""" |
|
175
|
|
|
if split_id == -1: |
|
176
|
|
|
if type(split_name) is tuple: |
|
177
|
|
|
time_array = [] |
|
178
|
|
|
for each_split in split_name: |
|
179
|
|
|
each_split_id = self.info['split_sets'].index(each_split) |
|
180
|
|
|
if 0 < each_split_id < len(self.info['split_sets']): |
|
181
|
|
|
time_array += self.info['split_timearray'][each_split_id] |
|
182
|
|
|
else: |
|
183
|
|
|
if split_name in self.info['split_sets']: |
|
184
|
|
|
split_id = self.info['split_sets'].index(split_name) |
|
185
|
|
|
if 0 < split_id < len(self.info['split_sets']): |
|
186
|
|
|
time_array = self.info['split_timearray'][split_id] |
|
187
|
|
|
else: |
|
188
|
|
|
logger.error('Failed to find split set with name %s.' % split_name) |
|
189
|
|
|
return None |
|
190
|
|
|
elif -1 < split_id < len(self.info['split_sets']): |
|
191
|
|
|
time_array = self.info['split_timearray'][split_id] |
|
192
|
|
|
else: |
|
193
|
|
|
logger.error('Split set index %d out of bound.' % split_id) |
|
194
|
|
|
return None |
|
195
|
|
|
return time_array |
|
196
|
|
|
|
|
197
|
|
|
@staticmethod |
|
198
|
|
|
def files_exist(dir_name): |
|
199
|
|
|
"""Check if the CASAS Fuel dataset files exist under dir_name |
|
200
|
|
|
""" |
|
201
|
|
|
data_filename = os.path.join(dir_name, 'data.hdf5') |
|
202
|
|
|
info_filename = os.path.join(dir_name, 'info.pkl') |
|
203
|
|
|
return os.path.isfile(data_filename) and os.path.isfile(info_filename) |
|
204
|
|
|
|
TinghuiWang /
pyActLearn
