forked from noxtoby/TADPOLE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
evalAllSubmissionsLive.py
464 lines (359 loc) · 17.1 KB
/
evalAllSubmissionsLive.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
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
import sys
import dropbox
from dropbox.files import WriteMode
from dropbox.exceptions import ApiError, AuthError
import argparse
import os
import evalOneSubmissionD4
import numpy as np
import pandas as pd
import csv
import string
import time
import datetime
import pickle
from scipy.stats import rankdata
# from configEval import *
from datetime import datetime
parser = argparse.ArgumentParser(usage='python3 leaderboardRunAll.py', description=r'''
Script evaluates new D4 entires and updates a live leaderboard table
Author: Razvan V. Marinescu, [email protected]
''')
parser.add_argument('--runPart', dest='runPart', default='RR',
help='which part of the script to run. Usually either LR or RR, where '
'LR means "load first part, run second part" while RR means run both parts')
parser.add_argument('--fast', dest='fast', type=int, default=1,
help='whether to run a fast version of the leaderboard.')
args = parser.parse_args()
TOKEN = open(os.path.expanduser('~/.dropboxTadpoleToken'), 'r').read()[:-1]
class DropboxObj:
def __init__(self):
self.TOKEN = TOKEN
self.dbx = self.createDropboxInstance()
def createDropboxInstance(self):
# Check for an access token
TOKEN = self.TOKEN
if (len(TOKEN) == 0):
sys.exit("ERROR: Looks like you didn't add your access token. "
"Open up backupuploadDropboxAPIv2.py in a text editor and "
"paste in your token in line 14.")
# Create an instance of a Dropbox class, which can make requests to the API.
print("Creating a Dropbox object...")
dbx = dropbox.Dropbox(TOKEN)
# Check that the access token is valid
try:
dbx.users_get_current_account()
except AuthError as err:
sys.exit("ERROR: Invalid access token; try re-generating an "
"access token from the app console on the web.")
return dbx
# Uploads contents of LOCALFILE to Dropbox
def upload(self, fullPathLocal, fullPathRemote):
print('fullPathRemote', fullPathRemote)
with open(fullPathLocal, 'rb') as f:
# We use WriteMode=overwrite to make sure that the settings in the file
# are changed on upload
print("Uploading " + fullPathLocal + " to Dropbox as " + fullPathRemote + "...")
try:
self.dbx.files_upload(f.read(), fullPathRemote, mode=WriteMode('overwrite'))
except ApiError as err:
# This checks for the specific error where a user doesn't have
# enough Dropbox space quota to upload this file
if (err.error.is_path() and
err.error.get_path().error.is_insufficient_space()):
sys.exit("ERROR: Cannot back up; insufficient space.")
elif err.user_message_text:
print(err.user_message_text)
sys.exit()
else:
print(err)
sys.exit()
# Download contents of LOCALFILE to Dropbox
def download(self, localPath, remotePath):
print("Downloading " + remotePath + " from Dropbox to " + localPath + " ...")
try:
self.dbx.files_download_to_file(localPath, remotePath)
except ApiError as err:
if err.user_message_text:
print(err.user_message_text)
sys.exit()
else:
print(err)
sys.exit()
def list_folder(self, folder, subfolder):
"""List a folder.
Return a dict mapping unicode filenames to
FileMetadata|FolderMetadata entries.
"""
path = '/%s/%s' % (folder, subfolder.replace(os.path.sep, '/'))
while '//' in path:
path = path.replace('//', '/')
path = path.rstrip('/')
try:
res = self.dbx.files_list_folder(path)
except dropbox.exceptions.ApiError as err:
print('Folder listing failed for', path, '-- assumed empty:', err)
return {}
else:
rv = {}
for entry in res.entries:
rv[entry.name] = entry
return rv
def formatStrRemoveNan(strFmt, n, replaceStr = '-'):
if np.isnan(n):
return replaceStr
else:
return strFmt % n
def writeHTMLtable(evalResults, htmlFile):
html = open(htmlFile, 'w')
## Add this manually in skeleton13.css in the dropbox folder under ProAD/public_html/css
manuallyAddToPageCssStyle = r'''
<style>
tr.d0 td {
background-color: #ffffff;
color: black;
max-width: 30px;
overflow: wrap;
text-overflow: ellipsis;
word-wrap: break-word;
}
tr.d1 td {
background-color: #ffffff;
color: black;
max-width: 30px;
overflow: wrap;
text-overflow: ellipsis;
word-wrap: break-word;
}
</style>
'''
text = 'Table last updated on %s. Update frequency: every 5 minutes' % (datetime.now().strftime('%Y-%m-%d %H:%M (BST +0)'))
text += '<table class="table table-dark table-striped table-hover table-sm dataTable sortable" style="align:center" >\n'
text += r'''
<thead>
<tr class="d4HeaderLive">
<th style="width:50px"> RANK<br></th>
<th style="width:80px"> FILE NAME</th>
<th style="width:50px"> MAUC RANK</th>
<th style="width:50px"> MAUC</th>
<th style="width:50px"> BCA</th>
<th style="width:50px"> ADAS RANK</th>
<th style="width:50px"> ADAS MAE</th>
<th style="width:50px"> ADAS WES</th>
<th style="width:50px"> ADAS CPA</th>
<th style="width:50px"> VENTS RANK</th>
<th style="width:50px"> VENTS MAE</th>
<th style="width:50px"> VENTS WES</th>
<th style="width:50px"> VENTS CPA</th>
<th style="width:80px"> DATE</th>
</tr>
</thead>
'''
colsToShow = ['RANK MAUC', 'MAUC', 'BCA',
'RANK ADAS', 'ADAS MAE', 'ADAS WES', 'ADAS CPA', 'RANK VENTS', 'VENTS MAE', 'VENTS WES', 'VENTS CPA']
text += '<tbody>'
evalResultsPerm = evalResults.loc[:, colsToShow]
formatStrsMeasures = ['%.1f', '%.3f', '%.3f', '%.1f', '%.2f', '%.2f', '%.2f', '%.1f', '%.2f', '%.2f', '%.2f', '%s']
# print(list(zip(formatStrsMeasures, evalResultsPerm.loc[0, :])))
# asda
for f in range(evalResults['MAUC'].shape[0]):
# text += '\n <tr class="d%d">' % (f % 2)
text += '\n <tr class="rowD4Live">'
text += '<td>%s</td>' % formatStrRemoveNan('%.1f',evalResults['RANK'].iloc[f], replaceStr='999')
text += '<td>%s</td><td>' % evalResults['FileName'].iloc[f]
text += '</td><td>'.join(
[formatStrRemoveNan(strFmt,n) for strFmt, n in zip(formatStrsMeasures, evalResultsPerm.loc[f, :])])
text += '<td class="rowD4LiveDate">%s</td>' % evalResults.loc[f, 'Date'].strftime('%Y-%m-%d %H:%M')
text += '</td></tr>\n'
text += '</tbody>\n</table>'
with open(htmlFile, "w") as f:
f.write(text)
def convRankToStr(rankVector):
unqRanks = np.unique(rankVector)
rankVectorStr = np.empty(rankVector.shape[0], object)
rankVectorStr[np.isnan(rankVector)] = '-'
for r in range(unqRanks.shape[0]):
rCurr = unqRanks[r]
if not np.isnan(rCurr):
idxCurrRank = np.where(np.abs(rankVector - rCurr) < 0.001)[0]
sizeRangeHalf = float(idxCurrRank.shape[0] - 1)/2
if sizeRangeHalf > 0.49:
rankVectorStr[idxCurrRank] = '%d-%d' % (round(rCurr-sizeRangeHalf), round(rCurr+sizeRangeHalf))
else:
rankVectorStr[idxCurrRank] = '%d' % rCurr
return rankVectorStr
def addOtherStatsTable(resTable):
notNanMaskMAUC = np.logical_not(np.isnan(np.array(resTable.loc[:, 'MAUC'].values.reshape(-1), float)))
notNanMaskADAS = np.logical_not(np.isnan(np.array(resTable.loc[:, 'ADAS MAE'].values.reshape(-1), float)))
notNanMaskVents = np.logical_not(np.isnan(np.array(resTable.loc[:, 'VENTS MAE'].values.reshape(-1), float)))
rankMAUC = np.nan * np.ones(resTable.shape[0])
rankADAS = np.nan * np.ones(resTable.shape[0])
rankVENTS = np.nan * np.ones(resTable.shape[0])
rankMAUC[notNanMaskMAUC] = rankdata(rankdata(-resTable.loc[:, 'MAUC'].values.reshape(-1)[notNanMaskMAUC],
method='average'), method='average')
rankADAS[notNanMaskADAS] = rankdata(rankdata(resTable.loc[:, 'ADAS MAE'].values.reshape(-1)[notNanMaskADAS], method='average'),
method='average')
rankVENTS[notNanMaskVents] = rankdata(rankdata(resTable.loc[:, 'VENTS MAE'].values.reshape(-1)[notNanMaskVents], method='average'),
method='average')
rankBCA = np.nan * np.ones(resTable.shape[0])
rankAdasWes = np.nan * np.ones(resTable.shape[0])
rankAdasCpa = np.nan * np.ones(resTable.shape[0])
rankVentsWes = np.nan * np.ones(resTable.shape[0])
rankVentsCpa = np.nan * np.ones(resTable.shape[0])
rankBCA[notNanMaskMAUC] = rankdata(rankdata(-resTable.loc[:, 'BCA'].values.reshape(-1)[notNanMaskMAUC],
method='average'), method='average')
rankAdasWes[notNanMaskADAS] = rankdata(rankdata(resTable.loc[:, 'ADAS WES'].values.reshape(-1)[notNanMaskADAS], method='average'), method='average')
rankAdasCpa[notNanMaskADAS] = rankdata(rankdata(resTable.loc[:, 'ADAS CPA'].values.reshape(-1)[notNanMaskADAS], method='average'), method='average')
rankVentsWes[notNanMaskVents] = rankdata(rankdata(resTable.loc[:, 'VENTS WES'].values.reshape(-1)[notNanMaskVents], method='average'), method='average')
rankVentsCpa[notNanMaskVents] = rankdata(rankdata(resTable.loc[:, 'VENTS CPA'].values.reshape(-1)[notNanMaskVents], method='average'), method='average')
rankOrder = np.nan * np.ones(resTable.shape[0])
rankSum = rankMAUC + rankADAS + rankVENTS
nnSumMask = np.logical_not(np.isnan(rankSum))
rankOrder[nnSumMask] = rankdata(rankSum[nnSumMask], method='average') # make them start from 1
print('rankOrder', rankOrder)
resTable.loc[:, 'RANK'] = rankOrder
resTable.loc[:, 'RANK MAUC'] = rankMAUC
resTable.loc[:, 'RANK ADAS'] = rankADAS
resTable.loc[:, 'RANK VENTS'] = rankVENTS
resTable.loc[:, 'RANK BCA'] = rankBCA
resTable.loc[:, 'RANK ADAS WES'] = rankAdasWes
resTable.loc[:, 'RANK ADAS CPA'] = rankAdasCpa
resTable.loc[:, 'RANK VENTS WES'] = rankVentsWes
resTable.loc[:, 'RANK VENTS CPA'] = rankVentsCpa
resTable.loc[:, 'RANK STR'] = convRankToStr(rankOrder)
resTable.loc[:, 'RANK MAUC STR'] = convRankToStr(rankMAUC)
resTable.loc[:, 'RANK ADAS STR'] = convRankToStr(rankADAS)
resTable.loc[:, 'RANK VENTS STR'] = convRankToStr(rankVENTS)
resTable.sort_values(by=['RANK', 'RANK MAUC'], ascending=True,inplace=True)
resTable.reset_index(drop=True, inplace=True)
resTable.RANK = resTable.RANK.astype(float)
# round the numbers for easy visualisation
roundDict = {'MAUC': 3, 'BCA': 3, 'ADAS MAE': 2, 'VENTS MAE': 4,
'ADAS WES': 2, 'VENTS WES': 4, 'ADAS CPA': 2, 'VENTS CPA': 2}
for c in roundDict.keys():
resTable[c] = resTable[c].astype(float).round(roundDict[c])
return resTable
def applyChangesDf(res, calcRandomBest=False):
''' add extra names to identify different submissions '''
res.loc[np.logical_and(res.FileName == 'EMC1', np.in1d(res.ID, [1, 5])), 'FileName'] = 'EMC1-Std'
res.loc[np.logical_and(res.FileName == 'EMC1', np.in1d(res.ID, [2,3,4,6,7,8])), 'FileName'] = 'EMC1-Custom'
res.loc[np.logical_and(res.FileName == 'DIKU-GeneralisedLog', np.in1d(res.ID, [1, 5])), 'FileName'] = 'DIKU-GeneralisedLog-Std'
res.loc[np.logical_and(res.FileName == 'DIKU-GeneralisedLog', np.in1d(res.ID, [3, 7])), 'FileName'] = 'DIKU-GeneralisedLog-Custom'
res.loc[np.logical_and(res.FileName == 'DIKU-ModifiedLog', np.in1d(res.ID, [1, 5])), 'FileName'] = 'DIKU-ModifiedLog-Std'
res.loc[np.logical_and(res.FileName == 'DIKU-ModifiedLog', np.in1d(res.ID, [3, 7])), 'FileName'] = 'DIKU-ModifiedLog-Custom'
res.loc[np.logical_and(res.FileName == 'DIKU-ModifiedMri', np.in1d(res.ID, [1, 5])), 'FileName'] = 'DIKU-ModifiedMri-Std'
res.loc[np.logical_and(res.FileName == 'DIKU-ModifiedMri', np.in1d(res.ID, [3, 7])), 'FileName'] = 'DIKU-ModifiedMri-Custom'
# scale ventricles to show them as percentage points
res.loc[:, 'VENTS MAE'] *= 100
res.loc[:, 'VENTS WES'] *= 100
return res
def getD2D3deepCopy(res):
resD2 = res[res.PredictionSet == 'D2'].copy(deep=True)
resD3 = res[np.in1d(res.PredictionSet, ['D3'])].copy(deep=True)
resDCustom = res[np.in1d(res.PredictionSet, ['Custom'])].copy(deep=True)
resD2.sort_values(by=['MAUC', 'BCA'], ascending=False, inplace=True)
resD2.reset_index(drop=True, inplace=True)
resD3.sort_values(by=['MAUC', 'BCA'], ascending=False, inplace=True)
resD3.reset_index(drop=True, inplace=True)
resDCustom.sort_values(by=['MAUC', 'BCA'], ascending=False, inplace=True)
resDCustom.reset_index(drop=True, inplace=True)
return resD2, resD3, resDCustom
def evalD4LeaderboardSubmissions(resDf, evalResFile, fileNameTag, predictionSet):
fileListAll = ldbDropbox.list_folder(uploadsFldRemote, '/')
fileNamesLong = [x for x in fileListAll.keys() if (x.startswith(fileNameTag) and x[-4:] == '.csv')]
fileNamesLong.sort()
print('fileNamesLong ', fileNamesLong)
os.system('mkdir -p %s' % submissionsFld)
nrEntries = len(fileNamesLong)
fileNamesShort = [f.split('.')[0][len(fileNameTag):] for f in fileNamesLong]
tableColumns = ('TeamName', 'PredictionSet', 'FileName', 'ID', 'RANK',
'RANK MAUC', 'RANK ADAS', 'RANK VENTS', 'MAUC', 'BCA', 'ADAS MAE',
'VENTS MAE', 'ADAS WES', 'VENTS WES', 'ADAS CPA', 'VENTS CPA',
'Comments', 'Date')
listSubIndToProc = []
for e, f in enumerate(fileNamesShort):
possibleIndexOfExistingEntry = \
np.where(np.logical_and(resDf['FileName'] == f, resDf['PredictionSet'] == predictionSet))[0]
if possibleIndexOfExistingEntry.shape[0] == 0:
listSubIndToProc += [e]
if len(listSubIndToProc) > 0:
nanSeries = pd.DataFrame(np.nan, index=range(len(listSubIndToProc)), columns=tableColumns)
nrEntriesSoFar = resDf.shape[0]
res = resDf.append(nanSeries, ignore_index=True)
d4Df = pd.read_csv(d4File)
d4Df['CognitiveAssessmentDate'] = [datetime.strptime(x, '%Y-%m-%d') for x in d4Df['CognitiveAssessmentDate']]
d4Df['ScanDate'] = [datetime.strptime(x, '%Y-%m-%d') for x in d4Df['ScanDate']]
mapping = {'CN': 0, 'MCI': 1, 'AD': 2}
d4Df.replace({'Diagnosis': mapping}, inplace=True)
nrEntriesSoFar = resDf.shape[0]
entryToAddIndex = nrEntriesSoFar
for f in listSubIndToProc:
fileNameLong = fileNamesLong[f]
fileNameShort = fileNamesShort[f]
# print('teamname ', fileNameShort)
remotePath = '%s/%s' % (uploadsFldRemote, fileNameLong)
localPath = '%s/%s' % (submissionsFld, fileNameLong)
ldbDropbox.download(localPath, remotePath)
metadataFileRemote = ldbDropbox.dbx.files_get_metadata(remotePath)
print('Evaluating %s' % fileNameLong)
forecastDf = pd.read_csv(localPath)
try:
resDf.loc[entryToAddIndex, 'TeamName'] = fileNamesShort[f]
resDf.loc[entryToAddIndex, 'FileName'] = fileNamesShort[f]
resDf.loc[entryToAddIndex, 'ID'] = 1
resDf.loc[entryToAddIndex, 'PredictionSet'] = predictionSet
resDf.loc[entryToAddIndex, 'Date'] = metadataFileRemote.server_modified
resDf.loc[entryToAddIndex, 'MAUC': 'VENTS CPA'] = \
evalOneSubmissionD4.evalOneSub(d4Df, forecastDf)
entryToAddIndex += 1
except :
print('Error while processing submission %s' % fileNameLong)
pass
# if not np.isnan(resDf['MAUC'].iloc[entryToAddIndex]):
# entryToAddIndex += 1
print(resDf)
return resDf
# else:
# dataStruct = pickle.load(open(evalResFile, 'rb'))
# fileDatesRemote = dataStruct['fileDatesRemote']
# resDf = dataStruct['res']
if __name__ == '__main__':
submissionsFld = 'd4LiveSubmissions'
evalResFile = '%s/evalResAllD4Live.npz' % submissionsFld
dropboxRemoteFolder = '/ProAD/public_html'
uploadsFldRemote = '/ProAD/uploads'
d4File = '../TADPOLE_D4_corr.csv'
ldbDropbox = DropboxObj()
if args.fast:
# load submissions already evaluated and only evaluate the new ones
dataStruct = pickle.load(open(evalResFile, 'rb'))
resDf = dataStruct['res']
else:
# res = pd.DataFrame(np.nan, index=range(nrEntries), columns=tableColumns)
dataStruct = pickle.load(open('%s/resTableJune19.npz' % submissionsFld, 'rb'))
resDf = dataStruct['res']
# print(resDf.columns)
# asda
idxToKeep = ~np.in1d(resDf['TeamName'], ['Consensus', 'Randomised', 'ATRI-Biostat'])
resDf = resDf.loc[idxToKeep, :]
resDf.reset_index(inplace=True)
resDf.loc[:,'Date'] = [ datetime.strptime('Jun 14 2019 2:00PM', '%b %d %Y %I:%M%p') for _ in range(resDf.shape[0])]
nrInitEntries = resDf.shape[0]
resDf = evalD4LeaderboardSubmissions(resDf, evalResFile, fileNameTag ='TADPOLE_Submission_D4Live_D2_', predictionSet='D2')
resDf = evalD4LeaderboardSubmissions(resDf, evalResFile, fileNameTag='TADPOLE_Submission_D4Live_D3_',
predictionSet='D3')
nrFinalEntries = resDf.shape[0]
if nrFinalEntries > nrInitEntries:
dataStruct = dict(res=resDf)
pickle.dump(dataStruct, open(evalResFile, 'wb'), protocol=pickle.HIGHEST_PROTOCOL)
resD2, resD3, _ = getD2D3deepCopy(resDf)
resAll = [resD2, resD3]
predictionSets = ['D2', 'D3']
for d in range(2):
addOtherStatsTable(resAll[d])
applyChangesDf(resAll[d])
htmlFile = 'D4TableLive%s.html' % predictionSets[d]
htmlFileFullPathRemote = '%s/%s' % (dropboxRemoteFolder, htmlFile)
htmlFileFullPathLocal = '%s/%s' % (submissionsFld, htmlFile)
writeHTMLtable(resAll[d], htmlFileFullPathLocal)
ldbDropbox.upload(htmlFileFullPathLocal, htmlFileFullPathRemote)