-
Notifications
You must be signed in to change notification settings - Fork 17
/
leaderboard.py
433 lines (377 loc) · 16.5 KB
/
leaderboard.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
from itertools import product
import numpy as np
import pandas as pd
from ..model.event import Event
from ..model.event import EventTeam
from ..model.submission import Submission
from ..model.team import Team
from .team import get_event_team_by_name
from .submission import get_bagged_scores
from .submission import get_scores
from .submission import get_submission_max_ram
from .submission import get_time
pd.set_option('display.max_colwidth', -1)
def _compute_leaderboard(session, submissions, leaderboard_type, event_name,
with_links=True):
"""Format the leaderboard.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
submissions : list of :class:`ramp_database.model.Submission`
The submission to report in the leaderboard.
leaderboard_type : {'public', 'private'}
The type of leaderboard to built.
event_name : str
The name of the event.
with_links : bool
Whether or not the submission name should be clickable.
Returns
-------
leaderboard : dataframe
The leaderboard in a dataframe format.
"""
record_score = []
event = session.query(Event).filter_by(name=event_name).one()
map_score_precision = {score_type.name: score_type.precision
for score_type in event.score_types}
for sub in submissions:
# take only max n bag
df_scores_bag = get_bagged_scores(session, sub.id)
highest_level = df_scores_bag.index.get_level_values('n_bag').max()
df_scores_bag = df_scores_bag.loc[(slice(None), highest_level), :]
df_scores_bag.index = df_scores_bag.index.droplevel('n_bag')
df_scores_bag = df_scores_bag.round(map_score_precision)
df_scores = get_scores(session, sub.id)
df_scores = df_scores.round(map_score_precision)
df_time = get_time(session, sub.id)
df_time = df_time.stack().to_frame()
df_time.index = df_time.index.set_names(['fold', 'step'])
df_time = df_time.rename(columns={0: 'time'})
df = pd.concat([df_scores, df_time], axis=1)
df_mean = df.groupby('step').mean()
df_std = df.groupby('step').std()
# select only the validation and testing steps and rename them to
# public and private
map_renaming = {'valid': 'public', 'test': 'private'}
df_mean = (df_mean.loc[list(map_renaming.keys())]
.rename(index=map_renaming)
.stack().to_frame().T)
df_std = (df_std.loc[list(map_renaming.keys())]
.rename(index=map_renaming)
.stack().to_frame().T)
df_scores_bag = (df_scores_bag.rename(index=map_renaming)
.stack().to_frame().T)
df = pd.concat([df_scores_bag, df_mean, df_std], axis=1,
keys=['bag', 'mean', 'std'])
df.columns = df.columns.set_names(['stat', 'set', 'score'])
# change the multi-index into a stacked index
df.columns = df.columns.map(lambda x: " ".join(x))
df['team'] = sub.team.name
df['submission'] = sub.name_with_link if with_links else sub.name
df['contributivity'] = int(round(100 * sub.contributivity))
df['historical contributivity'] = int(round(
100 * sub.historical_contributivity))
df['max RAM [MB]'] = get_submission_max_ram(session, sub.id)
df['submitted at (UTC)'] = pd.Timestamp(sub.submission_timestamp)
record_score.append(df)
# stack all the records
df = pd.concat(record_score, axis=0, ignore_index=True, sort=False)
# keep only second precision for the time stamp
df['submitted at (UTC)'] = df['submitted at (UTC)'].astype('datetime64[s]')
# rename the column of the time
df = df.rename(columns={'mean public time': 'train time [s]',
'std public time': 'train time std [s]',
'mean private time': 'test time [s]',
'std private time': 'test time std [s]'})
# reordered the column
stats_order = (['bag', 'mean', 'std'] if leaderboard_type == 'private'
else ['bag'])
dataset_order = (['public', 'private'] if leaderboard_type == 'private'
else ['public'])
score_order = ([event.official_score_name] +
[score_type.name for score_type in event.score_types
if score_type.name != event.official_score_name])
score_list = [
'{} {} {}'.format(stat, dataset, score)
for stat, dataset, score in product(stats_order, dataset_order,
score_order)
]
col_ordered = (
['team', 'submission'] +
score_list +
['contributivity', 'historical contributivity',
'train time [s]', 'test time [s]',
'max RAM [MB]', 'submitted at (UTC)']
)
df = df[col_ordered]
df = df.sort_values(
"bag {} {}".format(leaderboard_type, event.official_score_name),
ascending=event.get_official_score_type(session).is_lower_the_better
)
# rename the column name for the public leaderboard
if leaderboard_type == 'public':
df = df.rename(columns={
key: value for key, value in zip(score_list, score_order)
})
return df
def _compute_competition_leaderboard(session, submissions, leaderboard_type,
event_name):
"""Format the competition leaderboard.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
submissions : list of :class:`ramp_database.model.Submission`
The submission to report in the leaderboard.
leaderboard_type : {'public', 'private'}
The type of leaderboard to built.
event_name : str
The name of the event.
Returns
-------
competition_leaderboard : dataframe
The competition leaderboard in a dataframe format.
"""
event = session.query(Event).filter_by(name=event_name).one()
score_type = event.get_official_score_type(session)
score_name = event.official_score_name
private_leaderboard = _compute_leaderboard(session, submissions, 'private',
event_name, with_links=False)
col_selected_private = (['team', 'submission'] +
['bag private ' + score_name,
'bag public ' + score_name] +
['train time [s]', 'test time [s]',
'submitted at (UTC)'])
leaderboard_df = private_leaderboard[col_selected_private]
leaderboard_df = leaderboard_df.rename(
columns={'bag private ' + score_name: 'private ' + score_name,
'bag public ' + score_name: 'public ' + score_name}
)
# select best submission for each team
best_df = (leaderboard_df.groupby('team').min()
if score_type.is_lower_the_better
else leaderboard_df.groupby('team').max())
best_df = best_df[['public ' + score_name]].reset_index()
best_df['best'] = True
# merge to get a best indicator column then select best
leaderboard_df = pd.merge(
leaderboard_df, best_df, how='left',
left_on=['team', 'public ' + score_name],
right_on=['team', 'public ' + score_name]
)
leaderboard_df = leaderboard_df.fillna(False)
leaderboard_df = leaderboard_df[leaderboard_df['best']]
leaderboard_df = leaderboard_df.drop(columns='best')
# dealing with ties: we need the lowest timestamp
best_df = leaderboard_df.groupby('team').min()
best_df = best_df[['submitted at (UTC)']].reset_index()
best_df['best'] = True
leaderboard_df = pd.merge(
leaderboard_df, best_df, how='left',
left_on=['team', 'submitted at (UTC)'],
right_on=['team', 'submitted at (UTC)'])
leaderboard_df = leaderboard_df.fillna(False)
leaderboard_df = leaderboard_df[leaderboard_df['best']]
leaderboard_df = leaderboard_df.drop(columns='best')
# sort by public score then by submission timestamp, compute rank
leaderboard_df = leaderboard_df.sort_values(
by=['public ' + score_name, 'submitted at (UTC)'],
ascending=[score_type.is_lower_the_better, True])
leaderboard_df['public rank'] = np.arange(len(leaderboard_df)) + 1
# sort by private score then by submission timestamp, compute rank
leaderboard_df = leaderboard_df.sort_values(
by=['private ' + score_name, 'submitted at (UTC)'],
ascending=[score_type.is_lower_the_better, True])
leaderboard_df['private rank'] = np.arange(len(leaderboard_df)) + 1
leaderboard_df['move'] = \
leaderboard_df['public rank'] - leaderboard_df['private rank']
leaderboard_df['move'] = [
'{:+d}'.format(m) if m != 0 else '-' for m in leaderboard_df['move']]
col_selected = [
leaderboard_type + ' rank', 'team', 'submission',
leaderboard_type + ' ' + score_name, 'train time [s]', 'test time [s]',
'submitted at (UTC)'
]
if leaderboard_type == 'private':
col_selected.insert(1, 'move')
df = leaderboard_df[col_selected]
df = df.rename(columns={
leaderboard_type + ' ' + score_name: score_name,
leaderboard_type + ' rank': 'rank'
})
df = df.sort_values(by='rank')
return df
def get_leaderboard(session, leaderboard_type, event_name, user_name=None,
with_links=True):
"""Get a leaderboard.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
leaderboard_type : {'public', 'private', 'failed', 'new', \
'public competition', 'private competition'}
The type of leaderboard to generate.
event_name : str
The event name.
user_name : None or str, default is None
The user name. If None, scores from all users will be queried. This
parameter is discarded when requesting the competition leaderboard.
with_links : bool, default is True
Whether or not the submission name should be clickable.
Returns
-------
leaderboard : str
The leaderboard in HTML format.
"""
q = (session.query(Submission)
.filter(Event.id == EventTeam.event_id)
.filter(Team.id == EventTeam.team_id)
.filter(EventTeam.id == Submission.event_team_id)
.filter(Event.name == event_name))
if user_name is not None:
q = q.filter(Team.name == user_name)
submissions = q.all()
submission_filter = {'public': 'is_public_leaderboard',
'private': 'is_private_leaderboard',
'failed': 'is_error',
'new': 'is_new',
'public competition': 'is_in_competition',
'private competition': 'is_in_competition'}
submissions = [sub for sub in submissions
if (getattr(sub, submission_filter[leaderboard_type]) and
sub.is_not_sandbox)]
if not submissions:
return None
if leaderboard_type in ['public', 'private']:
df = _compute_leaderboard(
session, submissions, leaderboard_type, event_name,
with_links=with_links
)
elif leaderboard_type in ['new', 'failed']:
columns = ['team',
'submission',
'submitted at (UTC)']
if leaderboard_type == 'failed':
columns.append('error')
# we rely on the zip function ignore the submission state if the error
# column was not appended
data = [
{column: value
for column, value in zip(columns,
[sub.event_team.team.name,
sub.name_with_link,
pd.Timestamp(sub.submission_timestamp),
sub.state_with_link])}
for sub in submissions
]
df = pd.DataFrame(data, columns=columns)
else:
# make some extra filtering
submissions = [sub for sub in submissions if sub.is_public_leaderboard]
if not submissions:
return None
competition_type = ('public' if 'public' in leaderboard_type
else 'private')
df = _compute_competition_leaderboard(
session, submissions, competition_type, event_name
)
df_html = df.to_html(escape=False, index=False, max_cols=None,
max_rows=None, justify='left')
df_html = '<thead> {} </tbody>'.format(
df_html.split('<thead>')[1].split('</tbody>')[0]
)
return df_html
def update_leaderboards(session, event_name, new_only=False):
"""Update the leaderboards for a given event.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
event_name : str
The event name.
new_only : bool, default is False
Whether or not to update the whole leaderboards or only the new
submissions. You can turn this option to True when adding a new
submission in the database.
"""
event = session.query(Event).filter_by(name=event_name).one()
if not new_only:
event.private_leaderboard_html = get_leaderboard(
session, 'private', event_name
)
event.public_leaderboard_html_with_links = get_leaderboard(
session, 'public', event_name
)
event.public_leaderboard_html_no_links = get_leaderboard(
session, 'public', event_name, with_links=False
)
event.failed_leaderboard_html = get_leaderboard(
session, 'failed', event_name
)
event.public_competition_leaderboard_html = get_leaderboard(
session, 'public competition', event_name
)
event.private_competition_leaderboard_html = get_leaderboard(
session, 'private competition', event_name
)
event.new_leaderboard_html = get_leaderboard(
session, 'new', event_name
)
session.commit()
def update_user_leaderboards(session, event_name, user_name,
new_only=False):
"""Update the of a user leaderboards for a given event.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
event_name : str
The event name.
user_name : str
The user name. If None, scores from all users will be queried.
new_only : bool, default is False
Whether or not to update the whole leaderboards or only the new
submissions. You can turn this option to True when adding a new
submission in the database.
"""
event_team = get_event_team_by_name(session, event_name, user_name)
if not new_only:
event_team.leaderboard_html = get_leaderboard(
session, 'public', event_name, user_name
)
event_team.failed_leaderboard_html = get_leaderboard(
session, 'failed', event_name, user_name
)
event_team.new_leaderboard_html = get_leaderboard(
session, 'new', event_name, user_name
)
session.commit()
def update_all_user_leaderboards(session, event_name, new_only=False):
"""Update the leaderboards for all users for a given event.
Parameters
----------
session : :class:`sqlalchemy.orm.Session`
The session to directly perform the operation on the database.
event_name : str
The event name.
new_only : bool, default is False
Whether or not to update the whole leaderboards or only the new
submissions. You can turn this option to True when adding a new
submission in the database.
"""
event = session.query(Event).filter_by(name=event_name).one()
event_teams = session.query(EventTeam).filter_by(event=event).all()
for event_team in event_teams:
user_name = event_team.team.name
if not new_only:
event_team.leaderboard_html = get_leaderboard(
session, 'public', event_name, user_name
)
event_team.failed_leaderboard_html = get_leaderboard(
session, 'failed', event_name, user_name
)
event_team.new_leaderboard_html = get_leaderboard(
session, 'new', event_name, user_name
)
session.commit()