1
|
|
|
import sys |
2
|
|
|
import logging |
3
|
|
|
import numpy as np |
4
|
|
|
import matplotlib.pyplot as plt |
5
|
|
|
from matplotlib.ticker import MultipleLocator |
6
|
|
|
from matplotlib.patches import Rectangle |
7
|
|
|
|
8
|
|
|
logger = logging.getLogger(__name__) |
9
|
|
|
|
10
|
|
|
recall_scoring_labels = ['Correct', 'Fragmenting', 'Underfill-B', 'Underfill-E', 'Deletion'] |
11
|
|
|
fpr_scoring_labels = ['Correct', 'Merging', 'Overfill-B', 'Overfill-E', 'Insertion'] |
12
|
|
|
recall_scoring_indices = {'C': 0, 'D': 4, 'F': 1, 'U': 2, 'u': 3} |
13
|
|
|
fpr_scoring_indices = {'C': 0, 'I': 4, 'M': 1, 'O': 2, 'o': 3} |
14
|
|
|
|
15
|
|
|
|
16
|
|
|
def draw_per_class_recall(classes, class_colors, recall_array, filename=None): |
17
|
|
|
"""Draw recall array |
18
|
|
|
""" |
19
|
|
|
recall_np = np.empty((len(classes), len(recall_scoring_labels)), |
20
|
|
|
dtype=np.float) |
21
|
|
|
for i, row in enumerate(recall_array): |
22
|
|
|
for key in recall_scoring_indices: |
23
|
|
|
recall_np[i, recall_scoring_indices[key]] = row[key] |
24
|
|
|
recall_np /= np.sum(recall_np, axis=1, keepdims=True) |
25
|
|
|
|
26
|
|
|
ind = np.arange(len(classes)) |
27
|
|
|
width = 0.35 |
28
|
|
|
bottom = np.zeros((len(classes),)) |
29
|
|
|
bar_array = [] |
30
|
|
|
for i in range(len(recall_scoring_labels)): |
31
|
|
|
bar_array.append(plt.bar(ind, recall_np[:, i], width, |
32
|
|
|
alpha=(1-1/len(recall_scoring_labels) * i), |
33
|
|
|
color=class_colors, bottom=bottom)[0]) |
34
|
|
|
bottom += recall_np[:, i] |
35
|
|
|
plt.ylabel('Percentage') |
36
|
|
|
plt.xlabel('Classes') |
37
|
|
|
plt.xticks(ind, classes, rotation='vertical') |
38
|
|
|
plt.legend(bar_array, recall_scoring_labels) |
39
|
|
|
plt.show() |
40
|
|
|
|
41
|
|
|
|
42
|
|
|
def _get_bg_class_id(classes, background_class): |
43
|
|
|
# Verify Background Class first |
44
|
|
|
if background_class is not None: |
45
|
|
|
bg_class_id = classes.index(background_class) |
46
|
|
|
else: |
47
|
|
|
bg_class_id = -1 |
48
|
|
|
return bg_class_id |
49
|
|
|
|
50
|
|
|
|
51
|
|
|
def _get_metric_label_dict(metric_name='recall'): |
52
|
|
|
if metric_name == 'recall': |
53
|
|
|
metric_labels = recall_scoring_labels |
54
|
|
|
metric_indices = recall_scoring_indices |
55
|
|
|
else: |
56
|
|
|
metric_labels = fpr_scoring_labels |
57
|
|
|
metric_indices = fpr_scoring_indices |
58
|
|
|
return metric_labels, metric_indices |
59
|
|
|
|
60
|
|
|
|
61
|
|
|
def _gether_per_class_metrics(methods, classes, metric_arrays, as_percent, metric_labels, metric_indices): |
62
|
|
|
"""Prepare metrics for bar plot |
63
|
|
|
""" |
64
|
|
|
# Gather data for bar plot |
65
|
|
|
plot_metric_arrays = [] |
66
|
|
|
for j in range(len(methods)): |
67
|
|
|
cur_metric = np.empty((len(classes), len(metric_labels)), |
68
|
|
|
dtype=np.float) |
69
|
|
|
for i, row in enumerate(metric_arrays[j]): |
70
|
|
|
for key in metric_indices: |
71
|
|
|
cur_metric[i, metric_indices[key]] = row[key] |
72
|
|
|
# As percent |
73
|
|
|
if as_percent: |
74
|
|
|
cur_metric /= np.sum(cur_metric, axis=1, keepdims=True) |
75
|
|
|
# Append the metric for current methods |
76
|
|
|
plot_metric_arrays.append(cur_metric) |
77
|
|
|
return plot_metric_arrays |
78
|
|
|
|
79
|
|
|
|
80
|
|
|
def _compare_per_class_metrics(methods, classes, class_colors, metric_arrays, |
81
|
|
|
group_by='methods', filename=None, background_class=None, |
82
|
|
|
as_percent=True, metric_name='recall'): |
83
|
|
|
"""Compare per-class metrics between methods using bar-graph |
84
|
|
|
""" |
85
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name=metric_name) |
86
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
87
|
|
|
plot_metric_arrays = _gether_per_class_metrics(methods, classes, metric_arrays, as_percent, |
88
|
|
|
metric_labels, metric_indices) |
89
|
|
|
# Prepare Data and x-label |
90
|
|
|
xtick_labels = [] |
91
|
|
|
bar_colors = [] |
92
|
|
|
if bg_class_id < 0: |
93
|
|
|
plot_data = np.empty((len(methods) * len(classes), len(metric_labels))) |
94
|
|
|
else: |
95
|
|
|
plot_data = np.empty((len(methods) * (len(classes) - 1), len(metric_labels))) |
96
|
|
|
# Fill plot data with values |
97
|
|
|
if group_by == 'methods': |
98
|
|
|
num_base_axis = len(methods) |
99
|
|
|
if bg_class_id < 0: |
100
|
|
|
num_sec_axis = len(classes) |
101
|
|
|
else: |
102
|
|
|
num_sec_axis = len(classes) - 1 |
103
|
|
|
for j in range(len(classes)): |
104
|
|
|
if bg_class_id < 0 or j < bg_class_id: |
105
|
|
|
for i in range(len(methods)): |
106
|
|
|
bar_colors.append(class_colors[j]) |
107
|
|
|
xtick_labels.append(methods[i]) |
108
|
|
|
plot_data[j * num_base_axis + i, :] = plot_metric_arrays[i][j, :] |
109
|
|
|
elif j > bg_class_id: |
110
|
|
|
for i in range(len(methods)): |
111
|
|
|
bar_colors.append(class_colors[j]) |
112
|
|
|
xtick_labels.append(methods[i]) |
113
|
|
|
plot_data[(j-1) * num_base_axis + i, :] = plot_metric_arrays[i][j, :] |
114
|
|
|
else: |
115
|
|
|
if bg_class_id < 0: |
116
|
|
|
num_base_axis = len(classes) |
117
|
|
|
else: |
118
|
|
|
num_base_axis = len(classes) - 1 |
119
|
|
|
num_sec_axis = len(methods) |
120
|
|
|
for j in range(len(methods)): |
121
|
|
|
xtick_labels.append(methods[j]) |
122
|
|
|
for i in range(len(classes)): |
123
|
|
|
if bg_class_id < 0 or i < bg_class_id: |
124
|
|
|
bar_colors.append(class_colors[i]) |
125
|
|
|
plot_data[j * num_base_axis + i, :] = plot_metric_arrays[j][i, :] |
126
|
|
|
elif i > bg_class_id: |
127
|
|
|
bar_colors.append(class_colors[i]) |
128
|
|
|
plot_data[j * num_base_axis + i - 1, :] = plot_metric_arrays[j][i, :] |
129
|
|
|
# Calculate width and bar location |
130
|
|
|
width = 1/(num_base_axis + 1) |
131
|
|
|
ind = [] |
132
|
|
|
for i in range(num_sec_axis): |
133
|
|
|
for j in range(num_base_axis): |
134
|
|
|
ind.append(i + j * width + width) |
135
|
|
|
bottom = np.zeros((num_base_axis * num_sec_axis,)) |
136
|
|
|
# Set major and minor lines for y_axis |
137
|
|
|
if as_percent: |
138
|
|
|
minor_locator_value = 0.05 |
139
|
|
|
major_locator_value = 0.2 |
140
|
|
|
else: |
141
|
|
|
max_value = np.max(plot_data.sum(axis=1)) + 20 |
142
|
|
|
minor_locator_value = int(max_value/20) |
143
|
|
|
major_locator_value = int(max_value/5) |
144
|
|
|
# Set up x_label location |
145
|
|
|
xlabel_ind = [] |
146
|
|
|
if group_by == 'methods': |
147
|
|
|
xlabel_ind = [x + width/2 for x in ind] |
148
|
|
|
xlabel_rotation = 'vertical' |
149
|
|
|
else: |
150
|
|
|
xlabel_ind = [x + 0.5 for x in range(len(methods))] |
151
|
|
|
xlabel_rotation = 'horizontal' |
152
|
|
|
# Setup Figure |
153
|
|
|
fig, ax = plt.subplots() |
154
|
|
|
# Y-Axis |
155
|
|
|
minor_locator = MultipleLocator(minor_locator_value) |
156
|
|
|
major_locator = MultipleLocator(major_locator_value) |
157
|
|
|
ax.yaxis.set_minor_locator(minor_locator) |
158
|
|
|
ax.yaxis.set_major_locator(major_locator) |
159
|
|
|
ax.yaxis.grid(which="major", color='0.65', linestyle='-', linewidth=1) |
160
|
|
|
ax.yaxis.grid(which="minor", color='0.45', linestyle=' ', linewidth=1) |
161
|
|
|
# Plot Bar |
162
|
|
|
for i in range(len(metric_labels)): |
163
|
|
|
ax.bar(ind, plot_data[:, i], width, |
164
|
|
|
alpha=(1-1/len(metric_labels) * i), |
165
|
|
|
color=bar_colors, bottom=bottom) |
166
|
|
|
bottom += plot_data[:, i] |
167
|
|
|
if as_percent: |
168
|
|
|
plt.ylabel('Percentage') |
169
|
|
|
else: |
170
|
|
|
plt.ylabel('Count') |
171
|
|
|
plt.xlabel('Classes') |
172
|
|
|
plt.xticks(xlabel_ind, xtick_labels, rotation=xlabel_rotation, fontsize=6) |
173
|
|
|
# Prepare Legends |
174
|
|
|
patches = [] |
175
|
|
|
legend_labels = [] |
176
|
|
|
for i in range(len(metric_labels)): |
177
|
|
|
patches.append(Rectangle((0, 0), 0, 0, color='0.3', alpha=(1-1/len(metric_labels) * i))) |
178
|
|
|
legend_labels.append(metric_labels[i]) |
179
|
|
|
for i in range(len(classes)): |
180
|
|
|
if i == bg_class_id: |
181
|
|
|
continue |
182
|
|
|
patches.append(Rectangle((0, 0), 0, 0, color=class_colors[i])) |
183
|
|
|
legend_labels.append(classes[i]) |
184
|
|
|
plt.legend(patches, legend_labels, loc='center left', borderaxespad=0, bbox_to_anchor=(1.05, 0.5), |
185
|
|
|
prop={'size': 8}) |
186
|
|
|
plt.tight_layout() |
187
|
|
|
plt.title('Event-based Activity Analysis - %s' % metric_name) |
188
|
|
|
if filename is None: |
189
|
|
|
plt.show() |
190
|
|
|
else: |
191
|
|
|
plt.savefig(filename, bbox_inches='tight') |
192
|
|
|
|
193
|
|
|
|
194
|
|
|
def compare_per_class_recall(methods, classes, class_colors, recall_arrays, |
195
|
|
|
group_by='methods', filename=None, background_class=None, |
196
|
|
|
as_percent=True): |
197
|
|
|
"""Draw event.rst-based comparison between methods on Recall metric. |
198
|
|
|
|
199
|
|
|
Args: |
200
|
|
|
methods (:obj:`list` of :obj:`str`): List of names of different methods to be plotted. |
201
|
|
|
classes (:obj:`list` of :obj:`str`): List of target classes. |
202
|
|
|
class_colors (:obj:`list` of :obj:`str`): List of RGB color for corresponding classes in the ``classes`` list. |
203
|
|
|
recall_arrays (:obj:`list` of :obj:`numpy.ndarray`): List of recall arrays calculated for each methods. |
204
|
|
|
group_by (:obj:`str`): Group the bar graph of various 'methods' first or 'classes' first. Default to 'methods'. |
205
|
|
|
filename (:obj:`str`): The filename to save the plot. None if display on screen with pyplot. |
206
|
|
|
background_class (:obj:`str`): Background class. Usually it points to ``Other_Activity``. The statistics of |
207
|
|
|
background_class will be omitted from the plot. |
208
|
|
|
as_percent (:obj:`bool`): Whether or not to convert the accumulated value to percentage. |
209
|
|
|
""" |
210
|
|
|
_compare_per_class_metrics(methods, classes, class_colors, recall_arrays, |
211
|
|
|
group_by=group_by, filename=filename, background_class=background_class, |
212
|
|
|
as_percent=as_percent, metric_name='recall') |
213
|
|
|
|
214
|
|
|
|
215
|
|
|
def compare_per_class_precision(methods, classes, class_colors, precision_arrays, |
216
|
|
|
group_by='methods', filename=None, background_class=None, |
217
|
|
|
as_percent=True): |
218
|
|
|
"""Draw event.rst-based comparison between methods on precision metric. |
219
|
|
|
|
220
|
|
|
Args: |
221
|
|
|
methods (:obj:`list` of :obj:`str`): List of names of different methods to be plotted. |
222
|
|
|
classes (:obj:`list` of :obj:`str`): List of target classes. |
223
|
|
|
class_colors (:obj:`list` of :obj:`str`): List of RGB color for corresponding classes in the ``classes`` list. |
224
|
|
|
recall_arrays (:obj:`list` of :obj:`numpy.ndarray`): List of recall arrays calculated for each methods. |
225
|
|
|
group_by (:obj:`str`): Group the bar graph of various 'methods' first or 'classes' first. Default to 'methods'. |
226
|
|
|
filename (:obj:`str`): The filename to save the plot. None if display on screen with pyplot. |
227
|
|
|
background_class (:obj:`str`): Background class. Usually it points to ``Other_Activity``. The statistics of |
228
|
|
|
background_class will be omitted from the plot. |
229
|
|
|
as_percent (:obj:`bool`): Whether or not to convert the accumulated value to percentage. |
230
|
|
|
""" |
231
|
|
|
_compare_per_class_metrics(methods, classes, class_colors, precision_arrays, |
232
|
|
|
group_by=group_by, filename=filename, background_class=background_class, |
233
|
|
|
as_percent=as_percent, metric_name='precision') |
234
|
|
|
|
235
|
|
|
|
236
|
|
|
def draw_timeliness_hist(classes, class_colors, truth, prediction, truth_scoring, prediction_scoring, time_list, |
237
|
|
|
background_class): |
238
|
|
|
"""Get Timeliness Histogram for underfill and overfill |
239
|
|
|
""" |
240
|
|
|
start_mismatch, stop_mismatch = _get_timeliness_measures(classes, truth, prediction, |
241
|
|
|
time_list) |
242
|
|
|
bg_id = _get_bg_class_id(classes, background_class) |
243
|
|
|
num_classes = len(classes) |
244
|
|
|
# Plot histogram |
245
|
|
|
stack_to_plot = [] |
246
|
|
|
stack_of_colors = [] |
247
|
|
|
stack_of_labels = [] |
248
|
|
|
for i in range(num_classes): |
249
|
|
|
if i != bg_id: |
250
|
|
|
stack_to_plot.append(start_mismatch[i]) |
251
|
|
|
stack_of_colors.append(class_colors[i]) |
252
|
|
|
stack_of_labels.append(classes[i]) |
253
|
|
|
# Histo stack |
254
|
|
|
bins = np.linspace(-300, 300, 100) |
255
|
|
|
plt.figure() |
256
|
|
|
patches = [] |
257
|
|
|
for i in range(num_classes-1): |
258
|
|
|
patches.append(Rectangle((0, 0), 0, 0, color=stack_of_colors[i])) |
259
|
|
|
for i in range(num_classes-1): |
260
|
|
|
plt.subplot(num_classes-1, 1, i+1) |
261
|
|
|
plt.hist(stack_to_plot[i], bins=bins, alpha=0.7, color=stack_of_colors[i], label=stack_of_labels[i], lw=0) |
262
|
|
|
# plt.hist(stack_to_plot, bins=bins, alpha=0.7, color=stack_of_colors, label=stack_of_labels) |
263
|
|
|
plt.legend(patches, stack_of_labels, loc='center left', borderaxespad=0, bbox_to_anchor=(1.05, 0.5), |
264
|
|
|
prop={'size': 8}) |
265
|
|
|
plt.show() |
266
|
|
|
|
267
|
|
|
|
268
|
|
|
def _find_overlap_seg(seg_list, id): |
269
|
|
|
for seg_id in range(len(seg_list)): |
270
|
|
|
if seg_list[seg_id][1] < id: |
271
|
|
|
continue |
272
|
|
|
elif seg_list[seg_id][0] > id: |
273
|
|
|
return -1 |
274
|
|
|
else: |
275
|
|
|
return seg_id |
276
|
|
|
return -1 |
277
|
|
|
|
278
|
|
|
|
279
|
|
|
def _find_seg_start_within(seg_list, start, stop): |
280
|
|
|
for seg_id in range(len(seg_list)): |
281
|
|
|
if seg_list[seg_id][0] < start: |
282
|
|
|
continue |
283
|
|
|
elif seg_list[seg_id][0] > stop: |
284
|
|
|
return -1 |
285
|
|
|
else: |
286
|
|
|
return seg_id |
287
|
|
|
return -1 |
288
|
|
|
|
289
|
|
|
|
290
|
|
|
def _find_seg_end_within(seg_list, start, stop): |
291
|
|
|
found_seg_id = -1 |
292
|
|
|
for seg_id in range(len(seg_list)): |
293
|
|
|
if seg_list[seg_id][1] < start: |
294
|
|
|
continue |
295
|
|
|
elif seg_list[seg_id][0] > stop: |
296
|
|
|
return found_seg_id |
297
|
|
|
else: |
298
|
|
|
found_seg_id = seg_id |
299
|
|
|
return found_seg_id |
300
|
|
|
|
301
|
|
|
|
302
|
|
View Code Duplication |
def _get_timeoffset_measures(classes, truth, prediction, time_list): |
|
|
|
|
303
|
|
|
num_classes = len(classes) |
304
|
|
|
start_mismatch = [list([]) for i in range(num_classes)] |
305
|
|
|
stop_mismatch = [list([]) for i in range(num_classes)] |
306
|
|
|
# Processing segmentation first! |
307
|
|
|
for j in range(num_classes): |
308
|
|
|
pred_segs = [] |
309
|
|
|
truth_segs = [] |
310
|
|
|
prev_pred = False |
311
|
|
|
prev_truth = False |
312
|
|
|
tseg_start = 0 |
313
|
|
|
tseg_stop = 0 |
314
|
|
|
pseg_start = 0 |
315
|
|
|
pseg_stop = 0 |
316
|
|
|
for i in range(truth.shape[0]): |
317
|
|
|
cur_truth = (int(truth[i]) == j) |
318
|
|
|
cur_pred = (int(prediction[i]) == j) |
319
|
|
|
# Truth segments |
320
|
|
|
if cur_truth != prev_truth: |
321
|
|
|
if cur_truth: |
322
|
|
|
tseg_start = i |
323
|
|
|
elif tseg_stop != 0: |
324
|
|
|
truth_segs.append((tseg_start, tseg_stop)) |
325
|
|
|
tseg_stop = i |
326
|
|
|
# Prediction segments |
327
|
|
|
if cur_pred != prev_pred: |
328
|
|
|
if cur_pred: |
329
|
|
|
pseg_start = i |
330
|
|
|
elif pseg_stop != 0: |
331
|
|
|
pred_segs.append((pseg_start, pseg_stop)) |
332
|
|
|
pseg_stop = i |
333
|
|
|
prev_truth = cur_truth |
334
|
|
|
prev_pred = cur_pred |
335
|
|
|
# Add compensated segments to predictions egments |
336
|
|
|
for ts, (tseg_start, tseg_stop) in enumerate(truth_segs): |
337
|
|
|
ps = _find_overlap_seg(pred_segs, tseg_start) |
338
|
|
|
if ps == -1: |
339
|
|
|
# potential underfill or deletion |
340
|
|
|
ps = _find_seg_start_within(pred_segs, tseg_start, tseg_stop) |
341
|
|
|
if ps != -1: |
342
|
|
|
pseg_start = pred_segs[ps][0] |
343
|
|
|
offset = (time_list[tseg_start] - time_list[pseg_start]).total_seconds() |
344
|
|
|
if abs(offset) < 18000: |
345
|
|
|
start_mismatch[j].append(offset) |
346
|
|
|
else: |
347
|
|
|
pseg_start = pred_segs[ps][0] |
348
|
|
|
# Check the end of previous truth |
349
|
|
|
if ts > 1 and truth_segs[ts-1][1] >= pseg_start: |
350
|
|
|
continue |
351
|
|
|
else: |
352
|
|
|
offset = (time_list[tseg_start] - time_list[pseg_start]).total_seconds() |
353
|
|
|
if abs(offset) < 18000: |
354
|
|
|
# Calculate overfill |
355
|
|
|
start_mismatch[j].append((time_list[tseg_start] - time_list[pseg_start]).total_seconds()) |
356
|
|
|
for ts, (tseg_start, tseg_stop) in enumerate(truth_segs): |
357
|
|
|
ps = _find_overlap_seg(pred_segs, tseg_stop) |
358
|
|
|
if ps == -1: |
359
|
|
|
# potential underfill or deletion |
360
|
|
|
ps = _find_seg_end_within(pred_segs, tseg_start, tseg_stop) |
361
|
|
|
if ps != -1: |
362
|
|
|
pseg_stop = pred_segs[ps][1] |
363
|
|
|
offset = (time_list[tseg_stop] - time_list[pseg_stop]).total_seconds() |
364
|
|
|
if tseg_stop != pseg_stop and abs(offset) < 18000: |
365
|
|
|
stop_mismatch[j].append(offset) |
366
|
|
|
else: |
367
|
|
|
pseg_stop = pred_segs[ps][1] |
368
|
|
|
# Check the end of previous truth |
369
|
|
|
if ts < len(truth_segs) - 1 and truth_segs[ts-1][0] <= pseg_stop: |
370
|
|
|
continue |
371
|
|
|
else: |
372
|
|
|
offset = (time_list[tseg_stop] - time_list[pseg_stop]).total_seconds() |
373
|
|
|
if abs(offset) < 18000: |
374
|
|
|
# Calculate overfill |
375
|
|
|
stop_mismatch[j].append(offset) |
376
|
|
|
# print("class: %d" % j) |
377
|
|
|
# print("pred_segs: %d %s" % (len(pred_segs), str(pred_segs))) |
378
|
|
|
# print("truth_segs: %d %s" % (len(truth_segs), str(truth_segs))) |
379
|
|
|
# print("start_mismatch: %s" % start_mismatch) |
380
|
|
|
# print("stop_mismatch: %s" % stop_mismatch) |
381
|
|
|
return start_mismatch, stop_mismatch |
382
|
|
|
|
383
|
|
|
|
384
|
|
View Code Duplication |
def _get_timeliness_measures(classes, truth, prediction, time_list): |
|
|
|
|
385
|
|
|
num_classes = len(classes) |
386
|
|
|
start_mismatch = [list([]) for i in range(num_classes)] |
387
|
|
|
stop_mismatch = [list([]) for i in range(num_classes)] |
388
|
|
|
# Processing segmentation first! |
389
|
|
|
for j in range(num_classes): |
390
|
|
|
pred_segs = [] |
391
|
|
|
truth_segs = [] |
392
|
|
|
prev_pred = False |
393
|
|
|
prev_truth = False |
394
|
|
|
tseg_start = 0 |
395
|
|
|
tseg_stop = 0 |
396
|
|
|
pseg_start = 0 |
397
|
|
|
pseg_stop = 0 |
398
|
|
|
for i in range(truth.shape[0]): |
399
|
|
|
cur_truth = (int(truth[i]) == j) |
400
|
|
|
cur_pred = (int(prediction[i]) == j) |
401
|
|
|
# Truth segments |
402
|
|
|
if cur_truth != prev_truth: |
403
|
|
|
if cur_truth: |
404
|
|
|
tseg_start = i |
405
|
|
|
elif tseg_stop != 0: |
406
|
|
|
truth_segs.append((tseg_start, tseg_stop)) |
407
|
|
|
tseg_stop = i |
408
|
|
|
# Prediction segments |
409
|
|
|
if cur_pred != prev_pred: |
410
|
|
|
if cur_pred: |
411
|
|
|
pseg_start = i |
412
|
|
|
elif pseg_stop != 0: |
413
|
|
|
pred_segs.append((pseg_start, pseg_stop)) |
414
|
|
|
pseg_stop = i |
415
|
|
|
prev_truth = cur_truth |
416
|
|
|
prev_pred = cur_pred |
417
|
|
|
# Add compensated segments to predictions egments |
418
|
|
|
for ts, (tseg_start, tseg_stop) in enumerate(truth_segs): |
419
|
|
|
ps = _find_overlap_seg(pred_segs, tseg_start) |
420
|
|
|
if ps == -1: |
421
|
|
|
# potential underfill or deletion |
422
|
|
|
ps = _find_seg_start_within(pred_segs, tseg_start, tseg_stop) |
423
|
|
|
if ps != -1: |
424
|
|
|
pseg_start = pred_segs[ps][0] |
425
|
|
|
offset = (time_list[tseg_start] - time_list[pseg_start]).total_seconds() |
426
|
|
|
if tseg_start != pseg_start and abs(offset) < 18000: |
427
|
|
|
start_mismatch[j].append(offset) |
428
|
|
|
else: |
429
|
|
|
pseg_start = pred_segs[ps][0] |
430
|
|
|
# Check the end of previous truth |
431
|
|
|
if ts > 1 and truth_segs[ts-1][1] >= pseg_start: |
432
|
|
|
continue |
433
|
|
|
else: |
434
|
|
|
offset = (time_list[tseg_start] - time_list[pseg_start]).total_seconds() |
435
|
|
|
if tseg_start != pseg_start and abs(offset) < 18000: |
436
|
|
|
# Calculate overfill |
437
|
|
|
start_mismatch[j].append((time_list[tseg_start] - time_list[pseg_start]).total_seconds()) |
438
|
|
|
for ts, (tseg_start, tseg_stop) in enumerate(truth_segs): |
439
|
|
|
ps = _find_overlap_seg(pred_segs, tseg_stop) |
440
|
|
|
if ps == -1: |
441
|
|
|
# potential underfill or deletion |
442
|
|
|
ps = _find_seg_end_within(pred_segs, tseg_start, tseg_stop) |
443
|
|
|
if ps != -1: |
444
|
|
|
pseg_stop = pred_segs[ps][1] |
445
|
|
|
offset = (time_list[tseg_stop] - time_list[pseg_stop]).total_seconds() |
446
|
|
|
if tseg_stop != pseg_stop and abs(offset) < 18000: |
447
|
|
|
stop_mismatch[j].append(offset) |
448
|
|
|
else: |
449
|
|
|
pseg_stop = pred_segs[ps][1] |
450
|
|
|
# Check the end of previous truth |
451
|
|
|
if ts < len(truth_segs) - 1 and truth_segs[ts-1][0] <= pseg_stop: |
452
|
|
|
continue |
453
|
|
|
else: |
454
|
|
|
offset = (time_list[tseg_stop] - time_list[pseg_stop]).total_seconds() |
455
|
|
|
if tseg_stop != pseg_stop and abs(offset) < 18000: |
456
|
|
|
# Calculate overfill |
457
|
|
|
stop_mismatch[j].append(offset) |
458
|
|
|
# print("class: %d" % j) |
459
|
|
|
# print("pred_segs: %d %s" % (len(pred_segs), str(pred_segs))) |
460
|
|
|
# print("truth_segs: %d %s" % (len(truth_segs), str(truth_segs))) |
461
|
|
|
# print("start_mismatch: %s" % start_mismatch) |
462
|
|
|
# print("stop_mismatch: %s" % stop_mismatch) |
463
|
|
|
return start_mismatch, stop_mismatch |
464
|
|
|
|
465
|
|
|
|
466
|
|
|
def _get_timeliness_measures_depricated(classes, truth, prediction, truth_scoring, prediction_scoring, time_list): |
467
|
|
|
num_classes = len(classes) |
468
|
|
|
start_mismatch = [list([]) for i in range(num_classes)] |
469
|
|
|
stop_mismatch = [list([]) for i in range(num_classes)] |
470
|
|
|
# For each Underfill, Overfill |
471
|
|
|
prev_truth = -1 |
472
|
|
|
for i in range(truth.shape[0]): |
473
|
|
|
cur_truth = int(truth[i]) |
474
|
|
|
# Overfill/Underfill only occur at the boundary of any activity event, so look for the boundary first |
475
|
|
|
if cur_truth != prev_truth: |
476
|
|
|
truth_time = time_list[i] |
477
|
|
|
# Check the start boundary |
478
|
|
|
if truth[i] == prediction[i]: |
479
|
|
|
# If current prediction is correct, then it can only be overfill of current truth label. |
480
|
|
|
j = i - 1 |
481
|
|
|
while j >= 0 and prediction_scoring[j] == 'O': |
482
|
|
|
j -= 1 |
483
|
|
|
# If there is no overfill for cur_truth, and the current truth and prediction are the same, |
484
|
|
|
# then there is no start_boundary mismatch. |
485
|
|
|
start_mismatch[cur_truth].append((time_list[j + 1] - truth_time).total_seconds()) |
486
|
|
|
else: |
487
|
|
|
# If current prediction is incorrect, then it can only be underfill of current truth label at start |
488
|
|
|
# boundary. |
489
|
|
|
j = i |
490
|
|
|
while j < truth.shape[0] and truth_scoring[j] == 'U': |
491
|
|
|
j += 1 |
492
|
|
|
if j != i and j < truth.shape[0]: |
493
|
|
|
start_mismatch[cur_truth].append((time_list[j-1] - truth_time).total_seconds()) |
494
|
|
|
# Check the stop boundary |
495
|
|
|
if i > 0: |
496
|
|
|
if prediction[i-1] == truth[i-1]: |
497
|
|
|
# Previous prediction is correct, then it can only be overfill of previous truth. |
498
|
|
|
# If there is no overfill, the stop boundary is accurate |
499
|
|
|
j = i |
500
|
|
|
while prediction_scoring[j] == 'o': |
501
|
|
|
j += 1 |
502
|
|
|
stop_mismatch[prev_truth].append((time_list[j-1] - truth_time).total_seconds()) |
503
|
|
|
else: |
504
|
|
|
# Check Underfill for prev_truth (at the stop boundary) |
505
|
|
|
j = i - 1 |
506
|
|
|
while j >= 0 and truth_scoring[j] == 'u': |
507
|
|
|
j -= 1 |
508
|
|
|
if j != i - 1: |
509
|
|
|
stop_mismatch[prev_truth].append((time_list[j + 1] - truth_time).total_seconds()) |
510
|
|
|
if prev_truth != -1: |
511
|
|
|
if len(stop_mismatch[prev_truth]) > 0 and abs(stop_mismatch[prev_truth][-1]) > 1800: |
512
|
|
|
logger.warning('Stop mismatch is over half an hour: %s at %d (%s) - %f' % |
513
|
|
|
(classes[prev_truth], i, time_list[i], |
514
|
|
|
stop_mismatch[prev_truth][-1])) |
515
|
|
|
if len(start_mismatch[cur_truth]) > 0 and abs(start_mismatch[cur_truth][-1]) > 1800: |
516
|
|
|
logger.warning('Start mismatch is over half an hour: %s at %d (%s) - %f' % |
517
|
|
|
(classes[cur_truth], i, time_list[i], |
518
|
|
|
start_mismatch[cur_truth][-1])) |
519
|
|
|
# Update prev truth |
520
|
|
|
prev_truth = cur_truth |
521
|
|
|
# Sort all arrays |
522
|
|
|
for i in range(num_classes): |
523
|
|
|
start_mismatch[i].sort() |
524
|
|
|
stop_mismatch[i].sort() |
525
|
|
|
# Return |
526
|
|
|
return start_mismatch, stop_mismatch |
527
|
|
|
|
528
|
|
|
|
529
|
|
|
def generate_latex_table(methods, classes, recall_metrics, precision_matrics, |
530
|
|
|
background_class=None, filename=None, |
531
|
|
|
as_percent=True, metric_name='recall'): |
532
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
533
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name='recall') |
534
|
|
|
rmp = _gether_per_class_metrics(methods, classes, recall_metrics, True, |
535
|
|
|
metric_labels, metric_indices) |
536
|
|
|
rmr = _gether_per_class_metrics(methods, classes, recall_metrics, False, |
537
|
|
|
metric_labels, metric_indices) |
538
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name='precision') |
539
|
|
|
pmp = _gether_per_class_metrics(methods, classes, precision_matrics, True, |
540
|
|
|
metric_labels, metric_indices) |
541
|
|
|
pmr = _gether_per_class_metrics(methods, classes, precision_matrics, False, |
542
|
|
|
metric_labels, metric_indices) |
543
|
|
|
if filename is None: |
544
|
|
|
f = sys.stdout |
545
|
|
|
else: |
546
|
|
|
f = open(filename, 'w') |
547
|
|
|
f.write('\\multirow{2}{*}{Models} & \\multirow{2}{*}{Activities} & ' |
548
|
|
|
'\\multirow{2}{*}{Total Truth} & \\multicolumn{2}{|c|}{Recall} & ' |
549
|
|
|
'\\multirow{2}{*}{Total Prediction} & \\multicolumn{2}{|c|}{Precision} \\\\ \\hline\n') |
550
|
|
|
f.write('& & & C only & U included & & C only & O included \\\\ \\hline \n') |
551
|
|
|
for i, method in enumerate(methods): |
552
|
|
|
f.write('\\multirow{%d}{*}{%s} & ' % (len(classes), method.replace('_', '\_'))) |
553
|
|
|
for j, target in enumerate(classes): |
554
|
|
|
if j != 0: |
555
|
|
|
f.write('& ') |
556
|
|
|
f.write('%s & ' |
557
|
|
|
'%d & %d (%.2f) & %d (%.2f) & ' |
558
|
|
|
'%d & %d (%.2f) & %d (%.2f) \\\\ \n' % |
559
|
|
|
(target.replace('_', '\_'), |
560
|
|
|
rmr[i][j,:].sum(), rmr[i][j,0], rmp[i][j,0], |
561
|
|
|
rmr[i][j,0]+rmr[i][j,1]+rmr[i][j,2], rmp[i][j,0]+rmp[i][j,1]+rmp[i][j,2], |
562
|
|
|
pmr[i][j,:].sum(), pmr[i][j,0], pmp[i][j,0], |
563
|
|
|
pmr[i][j,0]+pmr[i][j,1]+pmr[i][j,2], pmp[i][j,0]+pmp[i][j,1]+pmp[i][j,2], |
564
|
|
|
) |
565
|
|
|
) |
566
|
|
|
f.write('\\hline\n') |
567
|
|
|
f.close() |
568
|
|
|
|
569
|
|
|
|
570
|
|
|
def generate_seg_latex_table(methods, classes, recall_metrics, precision_matrics, |
571
|
|
|
background_class=None, filename=None): |
572
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
573
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name='recall') |
574
|
|
|
rmp = _gether_per_class_metrics(methods, classes, recall_metrics, True, |
575
|
|
|
metric_labels, metric_indices) |
576
|
|
|
rmr = _gether_per_class_metrics(methods, classes, recall_metrics, False, |
577
|
|
|
metric_labels, metric_indices) |
578
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name='precision') |
579
|
|
|
pmp = _gether_per_class_metrics(methods, classes, precision_matrics, True, |
580
|
|
|
metric_labels, metric_indices) |
581
|
|
|
pmr = _gether_per_class_metrics(methods, classes, precision_matrics, False, |
582
|
|
|
metric_labels, metric_indices) |
583
|
|
|
if filename is None: |
584
|
|
|
f = sys.stdout |
585
|
|
|
else: |
586
|
|
|
f = open(filename, 'w') |
587
|
|
|
f.write('Metric & Activities') |
588
|
|
|
for method in methods: |
589
|
|
|
f.write('& %s' % method.replace('_', '\_')) |
590
|
|
|
f.write('\\\\ \\hline \n') |
591
|
|
|
for i, activity in enumerate(classes): |
592
|
|
View Code Duplication |
if i != bg_class_id: |
|
|
|
|
593
|
|
|
if i == 0: |
594
|
|
|
f.write('\multirow{%d}{*}{Recall} & ' % (len(classes) - 1)) |
595
|
|
|
else: |
596
|
|
|
f.write(' & ') |
597
|
|
|
f.write('%s ' % activity.replace('_', '\_')) |
598
|
|
|
# Find maximum and store index |
599
|
|
|
temp_array = np.array([rmp[j][i,0] for j in range(len(methods))]) |
600
|
|
|
max_index = temp_array.argpartition(-2)[-2:] |
601
|
|
|
for j, method in enumerate(methods): |
602
|
|
|
if j in max_index: |
603
|
|
|
f.write('& \\textbf{%d/%.2f\\%%} ' % (rmr[j][i,0], rmp[j][i,0]* 100)) |
604
|
|
|
else: |
605
|
|
|
f.write('& %d/%.2f\\%% ' % (rmr[j][i,0], rmp[j][i,0]* 100)) |
606
|
|
|
f.write('\\\\ \n') |
607
|
|
|
f.write('\\hline \n') |
608
|
|
|
for i, activity in enumerate(classes): |
609
|
|
View Code Duplication |
if i != bg_class_id: |
|
|
|
|
610
|
|
|
if i == 0: |
611
|
|
|
f.write('\multirow{%d}{*}{Precision} & ' % (len(classes) - 1)) |
612
|
|
|
else: |
613
|
|
|
f.write(' & ') |
614
|
|
|
f.write('%s ' % activity.replace('_', '\_')) |
615
|
|
|
# Find maximum and store index |
616
|
|
|
temp_array = np.array([pmp[j][i,0] for j in range(len(methods))]) |
617
|
|
|
max_index = temp_array.argpartition(-2)[-2:] |
618
|
|
|
for j, method in enumerate(methods): |
619
|
|
|
if j in max_index: |
620
|
|
|
f.write('& \\textbf{%d/%.2f\\%%} ' % (pmr[j][i,0], pmp[j][i,0]* 100)) |
621
|
|
|
else: |
622
|
|
|
f.write('& %d/%.2f\\%% ' % (pmr[j][i,0], pmp[j][i,0]* 100)) |
623
|
|
|
f.write('\\\\ \n') |
624
|
|
|
f.write('\\hline \n') |
625
|
|
|
|
626
|
|
|
|
627
|
|
|
def generate_event_recall_table(methods, classes, recall_metrics, |
628
|
|
|
background_class=None, filename=None): |
629
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
630
|
|
|
metric_labels, metric_indices = _get_metric_label_dict(metric_name='recall') |
631
|
|
|
rmp = _gether_per_class_metrics(methods, classes, recall_metrics, True, |
632
|
|
|
metric_labels, metric_indices) |
633
|
|
|
rmr = _gether_per_class_metrics(methods, classes, recall_metrics, False, |
634
|
|
|
metric_labels, metric_indices) |
635
|
|
|
if filename is None: |
636
|
|
|
f = sys.stdout |
637
|
|
|
else: |
638
|
|
|
f = open(filename, 'w') |
639
|
|
|
f.write('Activities') |
640
|
|
|
for method in methods: |
641
|
|
|
f.write('& %s' % method.replace('_', '\_')) |
642
|
|
|
f.write('\\\\ \\hline \n') |
643
|
|
|
for i, activity in enumerate(classes): |
644
|
|
|
if i != bg_class_id: |
645
|
|
|
f.write(' & ') |
646
|
|
|
f.write('%s ' % activity.replace('_', '\_')) |
647
|
|
|
# Find maximum and store index |
648
|
|
|
temp_array = np.array([rmp[j][i, 0] for j in range(len(methods))]) |
649
|
|
|
max_index = temp_array.argpartition(-2)[-2:] |
650
|
|
|
for j, method in enumerate(methods): |
651
|
|
|
if j in max_index: |
652
|
|
|
f.write('& \\textbf{%.2f\\%%} ' % (rmp[j][i,0]* 100)) |
653
|
|
|
else: |
654
|
|
|
f.write('& %.2f\\%% ' % (rmp[j][i,0]* 100)) |
655
|
|
|
f.write('\\\\ \n') |
656
|
|
|
f.write('\\hline \n') |
657
|
|
|
f.write('Recall (micro) &') |
658
|
|
|
total_correct = np.array([np.sum(rmr[j][:, 0]) - rmr[j][bg_class_id, 0] for j in range(len(methods))]) |
659
|
|
|
total_events = np.array([total_correct[j] + np.sum(rmr[j][:, 4]) - rmr[j][bg_class_id, 4] |
660
|
|
|
for j in range(len(methods))]) |
661
|
|
|
max_index = total_correct.argpartition(-2)[-2:] |
662
|
|
|
for j, method in enumerate(methods): |
663
|
|
|
if j in max_index: |
664
|
|
|
f.write('& \\textbf{%.2f\\%%} ' % (total_correct[j] / total_events[j] * 100)) |
665
|
|
|
else: |
666
|
|
|
f.write('& %.2f\\%% ' % (total_correct[j] / total_events[j] * 100)) |
667
|
|
|
f.write('\\\\ \n') |
668
|
|
|
f.write('\\hline \n') |
669
|
|
|
logger.debug('Total Events: %s' % str(total_events)) |
670
|
|
|
|
671
|
|
|
|
672
|
|
|
def generate_timeliness_table(methods, classes, result_array, |
673
|
|
|
background_class, filename=None): |
674
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
675
|
|
|
timeliness_values = [] |
676
|
|
|
for i, method in enumerate(methods): |
677
|
|
|
start_mismatch, stop_mismatch = _get_timeliness_measures(classes, result_array[i][0], result_array[i][1], |
678
|
|
|
result_array[i][4]) |
679
|
|
|
cur_timeliness = [start_mismatch[j] + stop_mismatch[j] for j in range(len(classes))] |
680
|
|
|
timeliness_values.append([np.abs(np.array(cur_timeliness[j])) for j in range(len(classes))]) |
681
|
|
|
# Average, <60, >60 |
682
|
|
|
if filename is None: |
683
|
|
|
f = sys.stdout |
684
|
|
|
else: |
685
|
|
|
f = open(filename, 'w') |
686
|
|
|
f.write('Activities & Metrics ') |
687
|
|
|
for method in methods: |
688
|
|
|
f.write('& %s' % method.replace('_', '\_')) |
689
|
|
|
f.write('\\\\ \\hline \n') |
690
|
|
|
for i, activity in enumerate(classes): |
691
|
|
|
if i != bg_class_id: |
692
|
|
|
f.write('\multirow{3}{*}{%s} & ' % activity.replace('_', '\_')) |
693
|
|
|
f.write('Average ') |
694
|
|
|
# Find maximum and store index |
695
|
|
|
for j, method in enumerate(methods): |
696
|
|
|
if len(timeliness_values[j][i]) == 0: |
697
|
|
|
average_time = 0. |
698
|
|
|
else: |
699
|
|
|
average_time = np.average(timeliness_values[j][i]) |
700
|
|
|
f.write('& %.2f s' % average_time) |
701
|
|
|
f.write('\\\\ \n') |
702
|
|
|
f.write(' & ') |
703
|
|
|
f.write('<60s ') |
704
|
|
|
for j, method in enumerate(methods): |
705
|
|
|
number = (timeliness_values[j][i] <= 60).sum() |
706
|
|
|
if len(timeliness_values[j][i]) == 0: |
707
|
|
|
percentage = 0.0 |
708
|
|
|
else: |
709
|
|
|
percentage = float(number)/len(timeliness_values[j][i]) * 100 |
710
|
|
|
f.write('& %d/%.2f\\%% ' % (number, percentage)) |
711
|
|
|
f.write('\\\\ \n') |
712
|
|
|
f.write(' & ') |
713
|
|
|
f.write('>60s ') |
714
|
|
|
for j, method in enumerate(methods): |
715
|
|
|
number = (timeliness_values[j][i] > 60).sum() |
716
|
|
|
if len(timeliness_values[j][i]) == 0: |
717
|
|
|
percentage = 0.0 |
718
|
|
|
else: |
719
|
|
|
percentage = float(number)/len(timeliness_values[j][i]) * 100 |
720
|
|
|
f.write('& %d/%.2f\\%% ' % (number, percentage)) |
721
|
|
|
f.write('\\\\ \\hline \n') |
722
|
|
|
|
723
|
|
|
|
724
|
|
|
def generate_timeliness_within60_table(methods, classes, result_array, |
725
|
|
|
background_class, filename=None): |
726
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
727
|
|
|
timeliness_values = [] |
728
|
|
|
for i, method in enumerate(methods): |
729
|
|
|
start_mismatch, stop_mismatch = _get_timeoffset_measures(classes, result_array[i][0], result_array[i][1], |
730
|
|
|
result_array[i][4]) |
731
|
|
|
cur_timeliness = [start_mismatch[j] + stop_mismatch[j] for j in range(len(classes))] |
732
|
|
|
timeliness_values.append([np.abs(np.array(cur_timeliness[j])) for j in range(len(classes))]) |
733
|
|
|
# Average, <60, >60 |
734
|
|
|
if filename is None: |
735
|
|
|
f = sys.stdout |
736
|
|
|
else: |
737
|
|
|
f = open(filename, 'w') |
738
|
|
|
f.write('\\textbf{Activities} ') |
739
|
|
|
for method in methods: |
740
|
|
|
f.write('& \\textbf{%s} ' % method.replace('_', ' ')) |
741
|
|
|
f.write('\\\\ \\midrule \n') |
742
|
|
|
for i, activity in enumerate(classes): |
743
|
|
|
if i != bg_class_id: |
744
|
|
|
f.write('%s & ' % activity.replace('_', ' ')) |
745
|
|
|
for j, method in enumerate(methods): |
746
|
|
|
number = (timeliness_values[j][i] <= 60).sum() |
747
|
|
|
if len(timeliness_values[j][i]) == 0: |
748
|
|
|
percentage = 0.0 |
749
|
|
|
else: |
750
|
|
|
percentage = float(number)/len(timeliness_values[j][i]) * 100 |
751
|
|
|
f.write('& %.2f\\%% ' % (percentage)) |
752
|
|
|
f.write('\\\\ \n') |
753
|
|
|
f.write('\\bottomrule\n') |
754
|
|
|
|
755
|
|
|
|
756
|
|
View Code Duplication |
def generate_timeliness_avg_table(methods, classes, result_array, |
|
|
|
|
757
|
|
|
background_class, filename=None): |
758
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
759
|
|
|
timeliness_values = [] |
760
|
|
|
for i, method in enumerate(methods): |
761
|
|
|
start_mismatch, stop_mismatch = _get_timeliness_measures(classes, result_array[i][0], result_array[i][1], |
762
|
|
|
result_array[i][4]) |
763
|
|
|
cur_timeliness = [start_mismatch[j] + stop_mismatch[j] for j in range(len(classes))] |
764
|
|
|
timeliness_values.append([np.abs(np.array(cur_timeliness[j])) for j in range(len(classes))]) |
765
|
|
|
# Average, <60, >60 |
766
|
|
|
if filename is None: |
767
|
|
|
f = sys.stdout |
768
|
|
|
else: |
769
|
|
|
f = open(filename, 'w') |
770
|
|
|
f.write('\\textbf{Activities} ') |
771
|
|
|
for method in methods: |
772
|
|
|
f.write('& \\textbf{%s} ' % method.replace('_', ' ')) |
773
|
|
|
f.write('\\\\ \\midrule \n') |
774
|
|
|
for i, activity in enumerate(classes): |
775
|
|
|
if i != bg_class_id: |
776
|
|
|
f.write('%s ' % activity.replace('_', ' ')) |
777
|
|
|
# Find maximum and store index |
778
|
|
|
for j, method in enumerate(methods): |
779
|
|
|
if len(timeliness_values[j][i]) == 0: |
780
|
|
|
average_time = 0. |
781
|
|
|
else: |
782
|
|
|
average_time = np.average(timeliness_values[j][i]) |
783
|
|
|
f.write('& %.1f' % average_time) |
784
|
|
|
f.write('\\\\ \n') |
785
|
|
|
f.write('\\bottomrule \n') |
786
|
|
|
|
787
|
|
|
|
788
|
|
View Code Duplication |
def generate_offset_per_table(methods, classes, result_array, |
|
|
|
|
789
|
|
|
background_class, filename=None): |
790
|
|
|
bg_class_id = _get_bg_class_id(classes, background_class) |
791
|
|
|
timeliness_values = [] |
792
|
|
|
for i, method in enumerate(methods): |
793
|
|
|
start_mismatch, stop_mismatch = _get_timeoffset_measures(classes, result_array[i][0], result_array[i][1], |
794
|
|
|
result_array[i][4]) |
795
|
|
|
cur_timeliness = [start_mismatch[j] + stop_mismatch[j] for j in range(len(classes))] |
796
|
|
|
timeliness_values.append([np.abs(np.array(cur_timeliness[j])) for j in range(len(classes))]) |
797
|
|
|
# Average, <60, >60 |
798
|
|
|
if filename is None: |
799
|
|
|
f = sys.stdout |
800
|
|
|
else: |
801
|
|
|
f = open(filename, 'w') |
802
|
|
|
f.write('\\textbf{Activities} ') |
803
|
|
|
for method in methods: |
804
|
|
|
f.write('& \\textbf{%s} ' % method.replace('_', ' ')) |
805
|
|
|
f.write('\\\\ \\midrule \n') |
806
|
|
|
for i, activity in enumerate(classes): |
807
|
|
|
if i != bg_class_id: |
808
|
|
|
f.write('%s ' % activity.replace('_', ' ')) |
809
|
|
|
# Find maximum and store index |
810
|
|
|
for j, method in enumerate(methods): |
811
|
|
|
total_num = len(timeliness_values[j][i])/2 |
812
|
|
|
nonzero_num = np.count_nonzero(timeliness_values[j][i]) |
813
|
|
|
f.write('& %d/%d' % (nonzero_num, total_num)) |
814
|
|
|
f.write('\\\\ \n') |
815
|
|
|
f.write('\\bottomrule \n') |
816
|
|
|
|
817
|
|
|
|