def SplitTrainTest(lSamples):
if config.get_bool('SPLIT_BY_EASY_HARD'):
setEasy = set(data.file_to_obj(config.get_string('EASY_CONNECTIONS_LIST_FILE')));
lTrain = filter(lambda samp: samp.pddlconn.sPddlTo in setEasy, lSamples);
lTest = filter(lambda samp: samp.pddlconn.sPddlTo not in setEasy, lSamples);
print "NUM Train:", len(lTrain), "Test:", len(lTest);
return lTrain, lTest;
if config.get_bool('SPLIT_BY_FIRST_30'):
setFirst30 = set(data.file_to_obj(config.get_string('FIRST_30_CONNECTIONS_LIST_FILE')));
lTrain = filter(lambda samp: samp.pddlconn.sPddlTo in setFirst30, lSamples);
lTest = filter(lambda samp: samp.pddlconn.sPddlTo not in setFirst30, lSamples);
print "NUM Train:", len(lTrain), "Test:", len(lTest);
return lTrain, lTest;
if config.get_bool('SPLIT_BY_SENTENCE'):
lRandomIndexes, iSplit = SplitSampleBySentence(lSamples);
else:
lRandomIndexes = range(len(lSamples));
random.shuffle(lRandomIndexes);
iSplit = len(lSamples)/2;
if config.get_bool('TRAIN_AND_TEST_ON_ALL'):
assert(not config.get_bool('TRAIN_ON_HALF_TEST_ON_ALL'));
lTrainIndexes = range(len(lSamples));
lTestIndexes = range(len(lSamples));
elif config.get_bool('TRAIN_ON_HALF_TEST_ON_ALL'):
lTrainIndexes = lRandomIndexes[:iSplit];
lTestIndexes = range(len(lSamples));
elif config.get_bool('TRAIN_ON_HALF_TEST_ON_HALF'):
lTrainIndexes = lRandomIndexes[:iSplit];
lTestIndexes = lRandomIndexes[iSplit:];
else:
assert False, 'No Test Train Split Method Specified';
lTrain = map(lambda x:lSamples[x], lTrainIndexes);
lTest = map(lambda x:lSamples[x], lTestIndexes);
return lTrain, lTest;
def LoadRewardsDict():
sFileRewards = config.get_string('REWARDS_FILE');
dConnRewardsIndexes = collections.defaultdict(lambda:{'iNumNeg':0, 'iNumPos':0});
iMaxIter = config.get_int('REWARDS_MAX_ITER');
iMinIter = config.get_int('REWARDS_MIN_ITER');
iCurIter = 0;
for sLine in open(sFileRewards):
sLine = sLine.strip();
lConns = sLine.split();
if len(lConns) > 5:
iCurIter += 1;
if iCurIter > iMaxIter:
break;
if iCurIter < iMinIter:
continue;
for sConn in lConns:
iFrom, iTo, iNumPos, iNumNeg = sConn.split(':');
iNumPos = int(iNumPos);
iNumNeg = int(iNumNeg);
iFrom = int(iFrom);
iTo = int(iTo);
dConnRewardsIndexes[(iFrom,iTo)]['iNumNeg'] += iNumNeg;
dConnRewardsIndexes[(iFrom,iTo)]['iNumPos'] += iNumPos;
dIndexToPredList = predicate.PredDictFileToPredListDict(config.get_string('PRED_DICT_FILE'), lambda x:x.iIndex);
dConnRewardsStrings = collections.defaultdict(lambda:{'iNumNeg':0, 'iNumPos':0});
for (iFrom, iTo), dPosNeg in dConnRewardsIndexes.items():
sFrom = dIndexToPredList[iFrom][0].GetObject();
sTo = dIndexToPredList[iTo][0].GetObject();
dConnRewardsStrings[(sFrom, sTo)]['iNumNeg'] += dPosNeg['iNumNeg'];
dConnRewardsStrings[(sFrom, sTo)]['iNumPos'] += dPosNeg['iNumPos'];
return dConnRewardsStrings;
def get_proxy_settings(config=None, conn=None):
r"""
Return proxy settings as a ProxySettings object
The caller must specify either config or conn.
Arguments:
- `config`: A osdlyrics.config.Config object, this object is used to retrive
proxy settings. If it is not set, the caller MUST set conn to a
valid D-Bus connection to create a Config object
- `conn`: A D-Bus connection object, this is used when `config` is not
specified.
"""
if config is None and conn is None:
raise ValueError('Either config or conn must be specified')
if config is None:
config = config.Config(conn)
proxy_type = config.get_string('Download/proxy')
if proxy_type.lower() == 'no':
return ProxySettings(protocol='no')
if proxy_type.lower() == 'manual':
protocol = config.get_string('Download/proxy-type')
host = config.get_string('Download/proxy-host')
port = config.get_int('Download/proxy-port')
username = config.get_string('Download/proxy-username')
passwd = config.get_string('Download/proxy-passwd')
return ProxySettings(protocol=protocol, host=host, port=port,
username=username, password=passwd)
if proxy_type.lower() == 'system':
return detect_system_proxy()
def CalcConnFileFScore():
print "Calcing FScore for:", config.get_string('CONN_FILE');
setConnTups = predicate.ReadConnectionsFileToTupSet();
setGoldConns = LoadGoldStringConnSet();
if config.get_bool('ANALYZE_ON_HARD'):
lEasy = data.file_to_obj(config.get_string('EASY_CONNECTIONS_LIST_FILE'));
iTruePos = 0;
iFalsePos = 0;
for sFrom, sTo in setConnTups:
if config.get_bool('ANALYZE_ON_HARD'):
if sTo in lEasy:
continue;
if (sFrom, sTo) in setGoldConns:
iTruePos += 1;
else:
iFalsePos += 1;
iFalseNeg = len(setGoldConns)-iTruePos;
if config.get_bool('ANALYZE_ON_HARD'):
iFalseNeg = len(setGoldConns)-iTruePos-len(lEasy);
iTrueNeg = 0;
fPrecision = float(iTruePos)/float(iTruePos+iFalsePos) if iTruePos > 0 else 0;
fRecall = float(iTruePos)/float(iTruePos+iFalseNeg) if iTruePos > 0 else 0;
fScore = 2*fPrecision*fRecall/(fPrecision+fRecall) if (fPrecision*fRecall) > 0 else 0;
print "TP:", iTruePos, "FP:", iFalsePos, "TN:", iTrueNeg, "FN:", iFalseNeg;
print "Precision:", fPrecision;
print "Recall:", fRecall;
print "FScore:", fScore;
def ReadConnectionsFileToTupSet():
dIndexToPred = PredDictFileToPredDict(config.get_string('PRED_DICT_FILE'), lambda x:x.iIndex);
setConnTups = set();
for sLine in open(config.get_string('CONN_FILE')):
sLine = sLine.strip();
lSplit = sLine.split('|');
setConnTups.add((dIndexToPred[int(lSplit[1])].GetObject(), dIndexToPred[int(lSplit[2])].GetObject()));
return setConnTups;
def __init__(self):
exchange_name = config.get_string('grid', 'name')
symbol = config.get_string('grid', 'symbol')
super(Grid, self).__init__(exchange_name, symbol)
self.base_line = config.get_float('grid', 'base_line') # 当前价格基准线
self.one_hand = config.get_float('grid', 'one_hand') # 一手买多少
logging.info("交易所：%s" % exchange_name)
logging.info("交易对：%s" % symbol)
logging.info("基线：%f" % self.base_line)
logging.info("一格：%f" % self.one_hand)
def Train(lSamples, j = 1, bSuppress = False):
sTrainFile = config.get_string('SVM_TRAIN');
sModelFile = config.get_string('SVM_MODEL');
Sample.WriteSvmFile(lSamples, sTrainFile);
if config.get_bool('USE_SVM_PERF'):
# version used for good svm results
iRetVal = PrintAndRun("../../bin/svm_perf_learn -c 10 -w 3 -l 1 -b 0 " + sTrainFile + " " + "model.svm", bSuppress = bSuppress);
#iRetVal = PrintAndRun("../../bin/svm_perf_learn -c 1 -w 3 -l 1 -b 0 " + sTrainFile + " " + "model.svm");
else:
iRetVal = PrintAndRun("../../bin/svm_learn -c 10 -b 0 -m 1000 -j " + str(j) + ' ' + sTrainFile + " " + "model.svm", bSuppress = bSuppress);
def WriteFirst30SvmConnectionsFile(lGranularSamples):
assert config.get_bool('SPLIT_BY_FIRST_30');
lTrainGranular, lTestGranular = SplitTrainTest(lGranularSamples);
lTestCollapsed, dFeatureWeights = TrainAndTestSvm(lTrainGranular, lTestGranular);
fScore, fPrec, fRecall = AnalyzePredsSimple(lTestCollapsed);
Sample.CollapsedSample.WriteConnections(lTestCollapsed, config.get_string('FIRST_30_SVM_CONNECTIONS_FILE'),
bAppend=False, bWritePredictions = True, bPosOnly = True);
# note that this one is train on train and test on train (yes those words are correct)
lTrainCollapsed, dFeatureWeights = TrainAndTestSvm(lTrainGranular, lTrainGranular);
fScore, fPrec, fRecall = AnalyzePredsSimple(lTrainCollapsed);
Sample.CollapsedSample.WriteConnections(lTrainCollapsed, config.get_string('FIRST_30_SVM_CONNECTIONS_FILE'),
bAppend=True, bWritePredictions = True, bPosOnly = True);
def ReadSentencesFromTextFile():
sSentenceFile = config.get_string('SENTENCE_FILE');
sOldSentenceFile = config.get_string('OLD_SENTENCE_FILE');
sDepFile = config.get_string('DEPENDANCY_FILE');
lSentences = ReadSentencesFromTextFileSimple(sSentenceFile);
if sDepFile != '':
Sentence.AddDepFeatures(lSentences, sDepFile);
if sOldSentenceFile != '':
lOldSentences = ReadSentencesFromTextFileSimple(sOldSentenceFile);
Sentence.AddOldSentenceFeatures(lSentences, lOldSentences);
return lSentences;
def Test(lSamples, bSuppress = False):
sTestFile = config.get_string('SVM_TEST');
sModelFile = config.get_string('SVM_MODEL');
sPredFile = config.get_string('SVM_PRED');
Sample.WriteSvmFile(lSamples, sTestFile);
if config.get_bool('USE_SVM_PERF'):
iRetVal = PrintAndRun("../../bin/svm_perf_classify " + sTestFile + " " + sModelFile + " " + sPredFile, bSuppress = bSuppress);
else:
iRetVal = PrintAndRun("../../bin/svm_classify " + sTestFile + " " + sModelFile + " " + sPredFile, bSuppress = bSuppress);
sPredLines = open(sPredFile).readlines();
lPreds = map(lambda x:float(x), sPredLines);
assert (len(lPreds) == len(lSamples));
for pred, sample in zip(lPreds, lSamples):
sample.fPred = float(pred);
def LoadPredStringSet():
lPreds = PredDictFileToPredList(config.get_string('PRED_DICT_FILE'));
setPreds = set();
for pred in lPreds:
setPreds.add(pred.GetObject());
print "NumPreds:", len(setPreds);
return setPreds;
def LoadGoldIndexesConnSet():
setGoldStringConns = LoadGoldStringConnSet();
dObjToPredList = predicate.PredDictFileToPredListDict(config.get_string('PRED_DICT_FILE'), lambda x:x.GetObject());
setGoldIndexConns = set();
for sFrom, sTo in setGoldStringConns:
for predFrom in dObjToPredList[sFrom]:
for predTo in dObjToPredList[sTo]:
setGoldIndexConns.add((predFrom.iIndex,predTo.iIndex));
return setGoldIndexConns;
def GenAllGranularSamplesFromList(lSentences, sLogFileName):
sSentenceLogFile = config.get_string('SENTENCE_LOG_FILE');
fLog = open(sSentenceLogFile, 'w');
lSamples = [];
iNumLoopy = 0;
for sentence in lSentences:
iCurNumLoopy, lCurSamples = sentence.GenAllGranularSamples(fLog);
lSamples.extend(lCurSamples);
iNumLoopy += iCurNumLoopy;
if iNumLoopy > 0:
print "NUM LOOPY:", iNumLoopy;
assert iNumLoopy < config.get_int('NUM_ALLOWED_LOOPY'), 'Too Many Loopy: ' + str(iNumLoopy) + ' NonLoopy: ' + str(len(lSamples));
sGoldDepFile = config.get_string('GOLD_DEP_FILE');
if sGoldDepFile != '':
dGoldDeps = data.file_to_obj_with_comments(sGoldDepFile);
# add the gold dep info
for sample in lSamples:
if (sample.pddlconn.sPddlTo in dGoldDeps) and (sample.pddlconn.sPddlFrom in dGoldDeps[sample.pddlconn.sPddlTo]):
sample.bGoldPos = True;
sPredDictFile = config.get_string('PRED_DICT_FILE');
if sPredDictFile != '':
lPredicates = predicate.PredDictFileToPredList(sPredDictFile);
dObjToPredList = collections.defaultdict(lambda:[]);
for predCur in lPredicates:
dObjToPredList[predCur.GetObject()].append(predCur);
for sample in lSamples:
sample.pddlconn.lFromPreds = dObjToPredList[sample.pddlconn.sPddlFrom];
sample.pddlconn.lToPreds = dObjToPredList[sample.pddlconn.sPddlTo];
else:
assert False;
#prune the unecessary features
dFeatureCounts = collections.defaultdict(lambda:0);
for sample in lSamples:
for iFeature in sample.features.GetFeatureIndexList():
dFeatureCounts[iFeature] += 1;
iMinFeatureCount = config.get_int('MIN_FEATURE_OCCURANCE_COUNT');
for sample in lSamples:
for iFeature in sample.features.GetFeatureIndexList():
if dFeatureCounts[iFeature] < iMinFeatureCount:
sample.features.RemoveFeature(iFeature);
return lSamples;
def LoadAllFullRewardsFiles():
sRewardsDir = config.get_string('FULL_REWARDS_LOG_DIR');
lRewards = [];
print "Loading 200 Files From:", sRewardsDir;
for iFileNum in range(config.get_int('REWARDS_MIN_ITER'), config.get_int('REWARDS_MAX_ITER')):
sFile = sRewardsDir + '/predictions.log.' + str(iFileNum);
lRewards.extend(LoadSingleFullRewardsFile(sFile));
print "Done Loading";
print "NumFF:", Reward.ComputeNumFfsFromList(lRewards);
sys.exit(-1);
return lRewards;
def LoadGoldStringConnSet():
sGoldDepFile = config.get_string('GOLD_DEP_FILE');
assert(sGoldDepFile != '');
dGoldDeps = data.file_to_obj_with_comments(sGoldDepFile);
setGoldConns = set();
for sTo, dFrom in dGoldDeps.items():
for sFrom in dFrom:
if sFrom == 'num':
continue;
setGoldConns.add((sFrom, sTo));
return setGoldConns;
iNumGold = len(setGoldConns);
def WriteBadPredCounts(dFalsePosCounts, dFalseNegCounts, dTruePosCounts, dTotalCounts):
if config.get_string('FALSE_POS_COUNTS_FILE') != '':
WriteCountsFile(dFalsePosCounts, dTotalCounts, config.get_string('FALSE_POS_COUNTS_FILE'));
if config.get_string('FALSE_NEG_COUNTS_FILE') != '':
WriteCountsFile(dFalseNegCounts, dTotalCounts, config.get_string('FALSE_NEG_COUNTS_FILE'));
if config.get_string('TRUE_POS_COUNTS_FILE') != '':
WriteCountsFile(dTruePosCounts, dTotalCounts, config.get_string('TRUE_POS_COUNTS_FILE'));
def Run():
if config.get_bool('CALC_CONN_FILE_FSCORE'):
CalcConnFileFScore();
return;
elif config.get_bool('CALC_EASY_HARD_CONNECTIONS'):
CalcEasyHardConnections();
return;
elif config.get_bool('LOAD_FULL_REWARDS'):
LoadFullRewards();
return;
lSentences = Sentence.ReadSentencesFromTextFile();
lGranularSamples = Sentence.GenAllGranularSamplesFromList(lSentences, 'sentences.log');
if config.get_bool('CALC_ALL_TEXT_FSCORE'):
CalcAllTextFScore(lGranularSamples);
elif config.get_string('GRANULAR_SAMPLE_FILE') != '':
Sample.GranularSample.WriteList(lGranularSamples, config.get_string('GRANULAR_SAMPLE_FILE'));
elif config.get_string('SENTENCES_AND_FEATURES_FILE') != '':
Sample.GranularSample.WriteDebugFromList(lGranularSamples, config.get_string('SENTENCES_AND_FEATURES_FILE'));
elif config.get_string('SVM_CONNECTIONS_FILE') != '':
WriteSvmConnectionsFile(lGranularSamples);
elif config.get_string('FIRST_30_SVM_CONNECTIONS_FILE') != '':
WriteFirst30SvmConnectionsFile(lGranularSamples);
elif config.get_string('COLLAPSED_MANUAL_TEXT_CONNECTIONS_FILE'):
lCollapsed = Sample.CollapseSamples(lGranularSamples);
Sample.CollapsedSample.WriteConnections(lCollapsed, config.get_string('COLLAPSED_MANUAL_TEXT_CONNECTIONS_FILE'), bPosOnly = True);
elif config.get_string('COLLAPSED_ALL_TEXT_CONNECTIONS_FILE'):
lCollapsed = Sample.CollapseSamples(lGranularSamples);
Sample.CollapsedSample.WriteConnections(lCollapsed, config.get_string('COLLAPSED_ALL_TEXT_CONNECTIONS_FILE'), bPosOnly = False);
elif config.get_bool('TRAIN_ON_REWARD_EVAL_ON_GOLD'):
TrainOnRewardEvalOnGold(lGranularSamples);
else:
Evaluate(lGranularSamples);
def TrainOnRewardEvalOnGold(lGranularSamples):
#lGranularSamples = lGranularSamples[:100];
assert(config.get_bool('COLLAPSE_FIRST'));
dConnRewards = LoadRewardsDict();
fNegMultiplier = PrintDebugInfo(dConnRewards);
#add the reward data to the samples themselves
lCollapsedSamples = Sample.CollapseSamples(lGranularSamples);
lNewSamples = [];
for sample in lCollapsedSamples:
dRewards = dConnRewards[sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo];
if dRewards['iNumPos'] > 0:
for iIter in range(dRewards['iNumPos']):
sampleNew = copy.copy(sample);
sampleNew._bPos = True;
lNewSamples.append(sampleNew);
else:
for iIter in range(int(math.ceil(dRewards['iNumNeg']*fNegMultiplier))):
sampleNew = copy.copy(sample);
sampleNew._bPos = False;
lNewSamples.append(sampleNew);
lCollapsedSamples = lNewSamples;
lFScores = [];
lPrecisions = [];
lRecalls = [];
lTrainCollapsed, lTestCollapsed = SplitTrainTest(lCollapsedSamples);
svm.Train(lTrainCollapsed);
svm.Test(lTestCollapsed);
# remove the duplicates
setAlreadySeen = set();
lTestNoDups = [];
for sample in lTestCollapsed:
tKey = (sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo);
if tKey not in setAlreadySeen:
setAlreadySeen.add(tKey);
lTestNoDups.append(sample);
fScore, fPrec, fRecall = AnalyzePredsSimple(lTestNoDups);
Sample.CollapsedSample.WriteConnections(lTestNoDups, config.get_string('SVM_REWARD_CONNECTIONS_FILE'),
bAppend=False, bWritePredictions = True, bPosOnly = True);
fOut = open('debug.txt', 'w');
for sample in lTestCollapsed:
print >> fOut, "Sample:", sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo, sample.fPred;
print "Precision: ", fPrec
print "Recall: ", fRecall
print "F-Score: ", fScore;
def LoadFullRewards():
dConnRewards = reward.LoadFullRewardsDict();
#print "Len:", len(dConnRewards);
#for (sFrom, sTo), cr in dConnRewards.items():
# print sFrom, sTo, "Pos:", cr.iNumPos, "NE:", cr.iNumEarlierNeg, "NR:", cr.iNumNoReachNeg;
#lGranularSamples = lGranularSamples[:100];
assert(config.get_bool('COLLAPSE_FIRST'));
#add the reward data to the samples themselves
lSentences = Sentence.ReadSentencesFromTextFile();
lGranularSamples = Sentence.GenAllGranularSamplesFromList(lSentences, 'sentences.log');
lCollapsedSamples = Sample.CollapseSamples(lGranularSamples);
setEasy = set(data.file_to_obj(config.get_string('EASY_CONNECTIONS_LIST_FILE')));
lTrainCollapsed = filter(lambda sample: sample.pddlconn.sPddlTo in setEasy, lCollapsedSamples);
lTestCollapsed = filter(lambda sample: sample.pddlconn.sPddlTo not in setEasy, lCollapsedSamples);
lTrainingSamples = GenerateTrainingSamplesFromRewards(dConnRewards, lTrainCollapsed);
if config.get_bool('SVM'):
svm.Train(lTrainingSamples);
svm.Test(lTestCollapsed);
elif config.get_bool('LOG_LINEAR'):
log_linear.TrainAndTestLogLinear(lTrainingSamples, lTestCollapsed);
else:
assert False;
# remove the duplicates
setAlreadySeen = set();
lTestNoDups = [];
for sample in lTestCollapsed:
tKey = (sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo);
if tKey not in setAlreadySeen:
setAlreadySeen.add(tKey);
lTestNoDups.append(sample);
fScore, fPrec, fRecall = AnalyzePredsSimple(lTestNoDups);
#Sample.CollapsedSample.WriteConnections(lTestNoDups, config.get_string('SVM_REWARD_CONNECTIONS_FILE'),
# bAppend=False, bWritePredictions = True, bPosOnly = True);
#fOut = open('debug.txt', 'w');
#for sample in lTestCollapsed:
# print >> fOut, "Sample:", sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo, sample.fPred;
print "Precision: ", fPrec
print "Recall: ", fRecall
print "F-Score: ", fScore;
def CalcAllTextFScore(lGranularSamples):
lCollapsedSamples = Sample.CollapseSamples(lGranularSamples);
lSorted = data.file_to_obj(config.get_string('SORTED_CONNECTIONS_LIST_FILE'));
lSorted.reverse();
lPos = [0 for i in range(len(lSorted))];
lNeg = [0 for i in range(len(lSorted))];
iPosTot = 0;
iNegTot = 0;
for sample in lCollapsedSamples:
if sample.bPos:
iPosTot += 1;
else:
iNegTot += 1;
for i in range(len(lSorted)):
if sample.pddlconn.sPddlTo != lSorted[i]:
continue;
if sample.bPos:
lPos[i] += 1;
else:
lNeg[i] += 1;
for i in range(len(lSorted)):
fPrecision = float(lPos[i])/float(lPos[i] + lNeg[i]) if (lPos[i] + lNeg[i]) != 0 else 0;
print lSorted[i], fPrecision;
print "Overall Precision:", float(iPosTot)/float(iNegTot);
# coding=utf-8
# 通知功能
import requests
import config
_DING_NOTIFY_URL = 'https://oapi.dingtalk.com/robot/send?access_token=' \
+ config.get_string('notify', 'ding_token')
def notify_by_dingding(msg):
"""
使用钉钉发送通知消息
:return:
"""
send_obj = {"msgtype": 'text', 'text': {'content': msg}}
requests.post(_DING_NOTIFY_URL, json=send_obj)
if __name__ == '__main__':
notify_by_dingding('币价涨了')
def AnalyzePredsSimple(lSamples):
if config.get_bool('FORCE_SINGLE_DIR'):
dSamples = {};
for sample in lSamples:
tKey = (sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo);
assert(tKey not in dSamples);
dSamples[tKey] = sample;
iNumTotal = 0;
iNumCorrect = 0;
iTruePos = 0;
iFalsePos = 0;
iTrueNeg = 0;
iFalseNeg = 0;
iThres = 0;
if config.get_bool('SVM'):
fThres = config.get_int('SVM_THRESHOLD');
elif config.get_bool('LOG_LINEAR'):
fThres = 0.5
else:
assert False;
if config.get_bool('CALC_FSCORE_ON_GOLD'):
setGoldStringConns = LoadGoldStringConnSet()
iNumGold = len(setGoldStringConns);
if config.get_bool('ANALYZE_ON_HARD'):
lEasy = data.file_to_obj(config.get_string('EASY_CONNECTIONS_LIST_FILE'));
fPredMin = sys.float_info.max;
fPredMax = -sys.float_info.max;
for sample in lSamples:
if config.get_bool('ANALYZE_ON_HARD'):
if sample.pddlconn.sPddlTo in lEasy:
continue;
if config.get_bool('TRAIN_ON_REWARD_EVAL_ON_GOLD'):
bActual = sample.GetGoldPos(bIgnoreDir = config.get_bool('IGNORE_DIR_FOR_EVAL'));
else:
bActual = sample.GetPos(bIgnoreDir = config.get_bool('IGNORE_DIR_FOR_EVAL'));
if config.get_bool('FORCE_SINGLE_DIR'):
fPred = sample.fPred;
tReverseKey = (sample.pddlconn.sPddlTo, sample.pddlconn.sPddlFrom);
fReversePred = dSamples[tReverseKey].fPred if tReverseKey in dSamples else -sys.maxint;
bNormalPred = (float(sample.fPred) > fThres);
bPred = ((float(sample.fPred) > fThres) and (float(fPred) >= float(fReversePred)));
if tReverseKey not in dSamples:
print "FORCE-MISSING";
elif (bNormalPred == bActual) and (bPred != bActual):
print "FORCE-BAD:", sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo, fPred, fReversePred;
elif (bNormalPred != bActual) and (bPred == bActual):
print "FORCE-GOOD:", sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo, fPred, fReversePred;
else:
print "FORCE-NEITHER:", sample.pddlconn.sPddlFrom, sample.pddlconn.sPddlTo, fPred, fReversePred;
else:
bPred = sample.GetPredPos(bIgnoreDir = config.get_bool('IGNORE_DIR_FOR_EVAL'));
fPredMin = min(fPredMin, sample.fPred);
fPredMax = max(fPredMax, sample.fPred);
iNumTotal += 1;
if bPred == bActual:
iNumCorrect += 1;
if bPred:
if bActual:
iTruePos += 1;
else:
iFalsePos += 1;
else:
if bActual:
iFalseNeg += 1;
else:
iTrueNeg += 1;
if config.get_bool('CALC_FSCORE_ON_GOLD'):
iFalseNeg = iNumGold - iTruePos;
if config.get_bool('ANALYZE_ON_HARD'):
iFalseNeg = iNumGold - iTruePos - len(lEasy);
fPrecision = float(iTruePos)/float(iTruePos+iFalsePos) if iTruePos > 0 else 0;
fRecall = float(iTruePos)/float(iTruePos+iFalseNeg) if iTruePos > 0 else 0;
fScore = 2*fPrecision*fRecall/(fPrecision+fRecall) if (fPrecision*fRecall) > 0 else 0;
print "FPred: min:", fPredMin, "max:", fPredMax;
print "FScore:", fScore, fPrecision, fRecall;
print "Frac Correct:", float(iNumCorrect)/float(iNumTotal), iNumCorrect, iNumTotal;
print "TP:", iTruePos, "FP:", iFalsePos, "TN:", iTrueNeg, "FN:", iFalseNeg;
print "FracPos:", float(iTruePos+iFalsePos)/float(iTrueNeg+iFalseNeg+iTruePos+iFalsePos);
return fScore, fPrecision, fRecall;
disp.print("Bat: %.2fV" % power.read_battery_voltage(),
posy=60,
fg=color.WHITE)
disp.update()
last_rx_time = 0
disp = display.open()
v_old = 0
pause = 1
interrupt.set_callback(interrupt.BLE, ble_callback)
interrupt.enable_callback(interrupt.BLE)
try:
vib_mode = int(config.get_string("exno_vib_mode"))
except:
pass
try:
led_mode = int(config.get_string("exno_led_mode"))
except:
pass
disp.clear()
disp.print(" Exp Notif", posy=0, fg=color.WHITE)
disp.print(" Logger", posy=20, fg=color.WHITE)
disp.print(" ", posy=40, fg=color.WHITE)
disp.print(" BL On ->", posy=60, fg=color.WHITE)
disp.update()
