def __init__(self, nParticipant, from_file = False):
self.nParticipant = nParticipant
self.events = EventLogger()
self.ratings = ParticipantRatings(self.nParticipant)
self.featureVectors = {} # {1:{'Fp1_theta': 2453476, 'Fp1_slow_alpha': 482418, ... , 'avgSkinRes': 69'}}
self.featureDF = pd.DataFrame # use special function
self.Y = {}
for trial in range(1,41):
self.featureVectors[trial] = {}
self.Y[trial] = 0
# seed and variables for splitting data
self.randomSeed = random.randint(1, 1000000)
self.X_train = {}
self.X_test = {}
self.X_validation = {}
self.Y_train = {}
self.Y_test = {}
self.Y_validation = {}
# if we not use precomputed feature from file - compute it from raw signals
if not from_file:
self.ratings = ParticipantRatings(self.nParticipant)
self.physSignalsFeatures = ParticipantSignalsFeatures(self.nParticipant)
self.physSignalsFeatures.computeFeatures(range(1, 41), range(1, 33), BANDS, FREQ, 0, 8063,
ASYM_ELECTRODE_PAIRS, ASYM_BANDS)
else:
self.loadFeatureVectorsFromCSV()
self.convertFeatureVectorsToDataFrame()
def initPanel(self):
_logger.info('initPanel')
self.addHandler()
g_statsCollector.eventHandlers += self.onEvent
g_statsCollector.start()
g_statsCollector.updateArenaInfo()
clientStatus = g_statsCollector.clientStatus
self.__panels = []
self.__keyHandlers = {}
for paneldef in self.__config.get('panelDefs', []):
if paneldef['channel'] == 'indicator':
panel = StatsIndicator(paneldef, clientStatus)
if 'events' in paneldef:
self.__eventHandlers += panel.onEvent
else:
self.__intervalHandlers += panel.update
if 'toggleKey' in paneldef['style']:
keyName = paneldef['style']['toggleKey']
keyId = getattr(Keys, keyName)
if keyId not in self.__keyHandlers:
self.__keyHandlers[keyId] = Event()
self.__keyHandlers[keyId] += panel.toggle
elif paneldef['channel'] == 'status':
panel = StatsLogger(paneldef, clientStatus)
self.__intervalHandlers += panel.update
elif paneldef['channel'] == 'event':
panel = EventLogger(paneldef, clientStatus)
self.__eventHandlers += panel.onEvent
self.__panels.append(panel)
session = dependency.instance(IBattleSessionProvider)
ctrl = session.shared.crosshair
self.changeView(ctrl.getViewID())
self.updateScreenPosition()
self.updateCrosshairPosition()
def __init__(self, stateDir, logDir, name):
self._name = name
self._logDir = logDir
ensureDirectory(logDir)
ensureDirectory(stateDir)
self._ids = Ids(stateDir, name)
self._invalidIds = Ids(stateDir, name + "_invalid")
self._state = State(stateDir, name)
self._eventlogger = EventLogger(logDir + '/' + name +'.events')
self._resetCounts()
def __init__(self, nParticipant, from_file = False):
self.nParticipant = nParticipant
self.events = EventLogger()
self.ratings = ParticipantRatings(self.nParticipant)
self.featureVectors = {} #
self.featureDF = pd.DataFrame # use special function
self.Y = {}
for trial in range(1, 41):
self.featureVectors[trial] = {}
self.Y[trial] = 0
# seed and variables for splitting data
self.randomSeed = random.randint(1, 1000000)
self.X_train = {}
self.X_test = {}
self.X_validation = {}
self.Y_train = {}
self.Y_test = {}
self.Y_validation = {}
# if we not use precomputed feature from file - compute it from raw signals
if not from_file:
self.featureVectors = self.unpackEEGDataToVectors()
else:
self.loadFeatureVectorsFromCSV()
self.convertFeatureVectorsToDataFrame()
class ParticipantFeatureVectors:
def __init__(self, nParticipant, from_file = False):
self.nParticipant = nParticipant
self.events = EventLogger()
self.ratings = ParticipantRatings(self.nParticipant)
self.featureVectors = {} # {1:{'Fp1_theta': 2453476, 'Fp1_slow_alpha': 482418, ... , 'avgSkinRes': 69'}}
self.featureDF = pd.DataFrame # use special function
self.Y = {}
for trial in range(1,41):
self.featureVectors[trial] = {}
self.Y[trial] = 0
# seed and variables for splitting data
self.randomSeed = random.randint(1, 1000000)
self.X_train = {}
self.X_test = {}
self.X_validation = {}
self.Y_train = {}
self.Y_test = {}
self.Y_validation = {}
# if we not use precomputed feature from file - compute it from raw signals
if not from_file:
self.ratings = ParticipantRatings(self.nParticipant)
self.physSignalsFeatures = ParticipantSignalsFeatures(self.nParticipant)
self.physSignalsFeatures.computeFeatures(range(1, 41), range(1, 33), BANDS, FREQ, 0, 8063,
ASYM_ELECTRODE_PAIRS, ASYM_BANDS)
else:
self.loadFeatureVectorsFromCSV()
self.convertFeatureVectorsToDataFrame()
def fillFeatureVectors(self):
self.addEEGSpectralToFeatureVector()
self.addEEGAsymetryToFeatureVector()
self.addGSRToFeatureVector()
def createYVector(self, yType = 'f'):
self.Y = {}
if yType == 'f':
if self.ratings.getFamiliarity() != []:
for trial in self.featureVectors.keys():
self.Y[trial] = self.ratings.familiarity.reset_index(drop=True).iloc[trial-1]
elif yType == 'a':
self.ratings.getArousal()
for trial in self.featureVectors.keys():
self.Y[trial] = self.ratings.arousal.reset_index(drop=True).iloc[trial - 1]
elif yType == 'v':
self.ratings.getValence()
for trial in self.featureVectors.keys():
self.Y[trial] = self.ratings.valence.reset_index(drop=True).iloc[trial - 1]
elif yType == 'l':
self.ratings.getLiking()
for trial in self.featureVectors.keys():
self.Y[trial] = self.ratings.liking.reset_index(drop=True).iloc[trial - 1]
elif yType == 'd':
self.ratings.getDominance()
for trial in self.featureVectors.keys():
self.Y[trial] = self.ratings.dominance.reset_index(drop=True).iloc[trial - 1]
elif yType == 'save':
for type in ['f', 'a', 'v', 'd', 'l']:
self.createYVector(yType=type)
self.saveYVectorToCSV(yType=type)
else:
print('No such Y vector type')
def convertFeatureVectorsToDataFrame(self):
self.featureDF = pd.DataFrame.from_dict(self.featureVectors, orient='index')
def addEEGSpectralToFeatureVector(self):
for trial in self.featureVectors.keys():
for electrode in range(len(EEG_CHANELS)):
for band in BANDS.keys():
feature_name = EEG_CHANELS[electrode] + '_' + band
self.featureVectors[trial][feature_name] = \
self.physSignalsFeatures.spectralEEGFeatures[trial][electrode+1][band]
def addEEGAsymetryToFeatureVector(self):
for trial in self.featureVectors.keys():
for electrodePair in ASYM_ELECTRODE_PAIRS:
leftE, rightE = electrodePair
for band in ASYM_BANDS.keys():
feature_name = EEG_CHANELS[leftE-1] + '-' + EEG_CHANELS[rightE-1] + '_' + band
self.featureVectors[trial][feature_name] = \
self.physSignalsFeatures.spectralEEGAsymetry[trial][leftE][band]
def addGSRToFeatureVector(self):
for trial in self.featureVectors.keys():
self.featureVectors[trial]['avgSkinRes'] = self.physSignalsFeatures.averageSkinResistance[trial]
def randomSplitSetForTraining(self, train=70, test=30, validation=0, seed=None):
'''
Split self.featureVectors and self.Y in random train, test and validation parts in a given proportions
:param train: proportion of train part, default 70
:param test: proportion of test part, default 30
:param validation: proportion of validation part, default 0
:param seed: seed for random, if None - there will be self.randomSeed used
:return: self.X_train, self.X_test, self.X_validation, self.Y_train, self.Y_test, self.Y_validation - feature
and target variable set divided in a given proportion
'''
# init and fill proportion variables
self.trainPart = train
self.testPart = test
self.validationPart = validation
# get random sample from feature vector index of test proportion length
if seed is None:
seed = self.randomSeed
random.seed(seed)
train_index = self.featureVectors.keys()
test_index = random.sample(train_index,
round(len(self.featureVectors.keys()) * test / (train + test + validation)))
test_index.sort() # to have ordered index
train_index = [item for item in train_index if item not in test_index]
# Not all model requires validation set, so we could skip it creation in such case
if validation != 0:
validation_index = random.sample(train_index, round(
len(self.featureVectors.keys()) * validation / (train + test + validation)))
validation_index.sort()
train_index = [item for item in train_index if item not in validation_index]
# create dict by created index
self.X_train = {key: self.featureVectors[key] for key in train_index}
try:
self.Y_train = {key: self.Y[key] for key in train_index}
except KeyError:
errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant),
'self.Y_train')
self.events.addEvent(204, errorMsg)
print(errorMsg)
self.X_test = {key: self.featureVectors[key] for key in test_index}
try:
self.Y_test = {key: self.Y[key] for key in test_index}
except KeyError:
errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant), 'self.Y_text')
self.events.addEvent(204, errorMsg)
print(errorMsg)
if validation != 0:
self.X_validation = {key: self.featureVectors[key] for key in validation_index}
try:
self.Y_validation = {key: self.Y[key] for key in validation_index}
except KeyError:
errorMsg = 'Participant {} self.Y is empty, so no data for {}'.format(str(self.nParticipant),
'self.Y_validation')
self.events.addEvent(204, errorMsg)
print(errorMsg)
return self.X_train, self.Y_train, self.X_test, self.Y_test, self.X_validation, self.Y_validation
def saveSplitedSetToCSV(self, seed=None):
if seed is None:
seed = self.randomSeed
names = ['X_train', 'X_test', 'X_validation', 'Y_train', 'Y_test', 'Y_validation']
sets = [self.X_train, self.X_test, self.X_validation, self.Y_train, self.Y_test, self.Y_validation]
pathname = 'training_data/seed={}&train={}&test={}&val={}/'.format(str(seed), str(self.trainPart),
str(self.testPart), str(self.validationPart))
if not os.path.isdir(pathname):
os.makedirs(pathname)
for name, set in zip(names, sets):
if set != {}:
file_name = '{1}_{2}_{0}.csv'.format(str(seed), str(self.nParticipant), name)
pd.DataFrame.from_dict(set, orient='index').to_csv(pathname+file_name)
def saveFeatureVectorToCSV(self):
filename = 'feature_vectors/FV{}.csv'.format(str(self.nParticipant))
pd.DataFrame.from_dict(self.featureVectors, orient='index').to_csv(filename)
def saveYVectorToCSV(self, yType):
filename = 'feature_vectors/YV'+str(self.nParticipant)+yType+'.csv'
pd.DataFrame.from_dict(self.Y, orient='index').to_csv(filename)
# in most cases we no need to recalculate features from data, so it's necessary to load the previously computed
# (and saved to *.csv) featureVectors
def loadFeatureVectorsFromCSV(self):
filename = 'feature_vectors/FV{}.csv'.format(str(self.nParticipant))
self.featureVectors = pd.DataFrame.from_csv(filename).to_dict(orient='index')
class HarvesterLog(object):
def __init__(self, stateDir, logDir, name):
self._name = name
self._logDir = logDir
ensureDirectory(logDir)
ensureDirectory(stateDir)
self._ids = Ids(stateDir, name)
self._invalidIds = Ids(stateDir, name + "_invalid")
self._state = State(stateDir, name)
self._eventlogger = EventLogger(logDir + '/' + name +'.events')
self._resetCounts()
def isCurrentDay(self, date):
return date.split('T')[0] == self._state.getTime().split()[0]
def startRepository(self):
self._resetCounts()
self._state.markStarted()
def _resetCounts(self):
self._harvestedCount = 0
self._uploadedCount = 0
self._deletedCount = 0
def totalIds(self):
return len(self._ids)
def totalInvalidIds(self):
return len(self._invalidIds)
def eventLogger(self):
# Should be removed, but is still used in Harvester.
return self._eventlogger
def markDeleted(self):
self._ids.clear()
self._state.markDeleted()
self._eventlogger.logSuccess('Harvested/Uploaded/Deleted/Total: 0/0/0/0, Done: Deleted all ids.', id=self._name)
def endRepository(self, token, responseDate):
self._state.markHarvested(self.countsSummary(), token, responseDate)
self._eventlogger.logSuccess('Harvested/Uploaded/Deleted/Total: %s, ResumptionToken: %s' % (self.countsSummary(), token), id=self._name)
def endWithException(self, exType, exValue, exTb):
self._state.markException(exType, exValue, self.countsSummary())
error = '|'.join(str.strip(s) for s in traceback.format_exception(exType, exValue, exTb))
self._eventlogger.logError(error, id=self._name)
def countsSummary(self):
return '%d/%d/%d/%d' % (self._harvestedCount, self._uploadedCount, self._deletedCount, self.totalIds())
def close(self):
self._eventlogger.close()
self._ids.close()
self._invalidIds.close()
self._state.close()
def notifyHarvestedRecord(self, uploadid):
self._removeFromInvalidData(uploadid)
self._harvestedCount += 1
def uploadIdentifier(self, uploadid):
self._ids.add(uploadid)
self._uploadedCount += 1
def deleteIdentifier(self, uploadid):
self._ids.remove(uploadid)
self._deletedCount += 1
def logInvalidData(self, uploadid, message):
self._invalidIds.add(uploadid)
filePath = self._invalidDataMessageFilePath(uploadid)
ensureDirectory(dirname(filePath))
open(filePath, 'w').write(message)
def logIgnoredIdentifierWarning(self, uploadid):
self._eventlogger.logWarning('IGNORED', uploadid)
def clearInvalidData(self, repositoryId):
for id in list(self._invalidIds):
if id.startswith("%s:" % repositoryId):
self._invalidIds.remove(id)
rmtree(join(self._logDir, INVALID_DATA_MESSAGES_DIR, repositoryId))
def hasWork(self, continuousInterval=None):
if continuousInterval is not None:
from_ = self._state.from_
if from_ and 'T' not in from_:
from_ += "T00:00:00Z"
return from_ is None or ZuluTime().epoch - ZuluTime(from_).epoch > continuousInterval
return self._state.token or self._state.from_ is None or not self.isCurrentDay(self._state.from_)
def state(self):
return self._state
def invalidIds(self):
return list(self._invalidIds)
def _removeFromInvalidData(self, uploadid):
self._invalidIds.remove(uploadid)
invalidDataMessageFilePath = self._invalidDataMessageFilePath(uploadid)
if isfile(invalidDataMessageFilePath):
remove(invalidDataMessageFilePath)
def _invalidDataMessageFilePath(self, uploadid):
repositoryId, recordId = uploadid.split(":", 1)
return join(self._logDir, INVALID_DATA_MESSAGES_DIR, escapeFilename(repositoryId), escapeFilename(recordId))