def evaluateSpecificDataSet(eng):
# evaluate data set
dRectWnd = 2.0
dSMWnd = 2.0
dSCWnd = 0.15
strCoder = ecc.CODER_RS
m = 4
n = 2**m-1
k = 1
r = (n-k)/2
nInterleaving = 25
print "%s: n=%d, k=%d, m=%d, r=%d, interleave=%d" % \
(strCoder, n, k, m, r, nInterleaving)
strWorkingDir = "../../data/evaluation/selected_set/"
lsFilePath = cf.getFileList(strWorkingDir, None)
strKeyOutput = "keys"
with open(strWorkingDir+strKeyOutput, 'w+') as hFile:
srMean, srStd, dfDetailed = evaluateDataSet('selected',
strWorkingDir, lsFilePath,
dRectWnd, dSMWnd, dSCWnd,
eng, strCoder, n, k, m, r,
nInterleaving, bOutputData=False,
bReconciliation=False,
hKeyOutput=hFile)
return srMean, srStd, dfDetailed
def evaluateMutualInformation():
"""
scatter plot & mutual information
"""
strWorkingDir = "../../data/evaluation/mutual_info/"
lsFilePath = cf.getFileList(strWorkingDir, None)
lsOutputData = []
srMean, srStd, dfDetailed = evaluateDataSet('selected',
strWorkingDir, lsFilePath,
bSourceEncoding = True,
bReconciliation = False,
bOutputData=True,
lsOutputData=lsOutputData)
dfData = pd.concat(lsOutputData, axis=0)
return dfData, lsOutputData
def evaluateGesture(eng):
"""
This function evaluates the performance of different gestures.
"""
# parameters
dRectWnd = 2.0
dSMWnd = 2.0
dSCWnd = 0.1
strCoder = ecc.CODER_GOLAY
m = 1
n = 23
k = 12
r = 3
nInterleaving = 25
print "%s: n=%d, k=%d, m=%d, r=%d, interleave=%d" % \
(strCoder, n, k, m, r, nInterleaving)
lsResult = []
strWorkingDir = "../../data/evaluation/gesture/"
for strLabel in ['g1', 'g2', 'g3']:
strFileNamePattern= strLabel
lsFilePath = cf.getFileList(strWorkingDir, strFileNamePattern)
if (len(lsFilePath) != 0 ):
srMean, srStd, dfDetailed = evaluateDataSet(strLabel,
strWorkingDir, lsFilePath,
dRectWnd, dSMWnd, dSCWnd,
eng, strCoder, n, k, m, r,
nInterleaving)
lsResult.append(srMean)
# print out rotten apples
dfSelected = dfDetailed[dfDetailed[sd.BER_USER_SRC]>=0.1]
if(dfSelected.shape[0] != 0):
print "--records with high BER--"
print dfSelected[\
[sd.FILE_NAME, sd.BER_USER_SRC, sd.BER_USER_EC]]
print "----\n"
dfSummary = pd.concat(lsResult, axis=1)
return dfSummary
def evaluateDistance(eng):
"""
This function evaluate the effect of distance btw A and B
"""
# parameters
dRectWnd = 2.0
dSMWnd = 2.0
dSCWnd = 0.1
strCoder = ecc.CODER_GOLAY
m = 1
n = 23
k = 12
r = int(math.floor((n-k)/2.0) )
nInterleaving = 25
print "%s: n=%d, k=%d, m=%d, r=%d, interleave=%d" % \
(strCoder, n, k, m, r, nInterleaving)
lsResult = []
strWorkingDir = "../../data/evaluation/distance/"
for strLabel in ['d1', 'd2', 'd3']:
strFileNamePattern= strLabel
lsFilePath = cf.getFileList(strWorkingDir, strFileNamePattern)
srMean, srStd, dfDetailed = evaluateDataSet(strLabel,
strWorkingDir, lsFilePath,
dRectWnd, dSMWnd, dSCWnd,
eng, strCoder, n, k, m, r,
nInterleaving)
lsResult.append(srMean)
# print out rotten apples
dfSelected = dfDetailed[dfDetailed[sd.BER_USER_SRC]>=0.1]
if(dfSelected.shape[0] != 0):
print "--records with high BER--"
print dfSelected[\
[sd.FILE_NAME, sd.BER_USER_SRC, sd.BER_USER_EC]]
print "----\n"
dfSummary = pd.concat(lsResult, axis=1)
return dfSummary
def evaluateShapeCodingParams(eng):
"""
This function evaluate the parameter of shape coding
"""
# select data
strWorkingDir = "../../data/evaluation/BER/"
strFileNamePattern= None
lsFilePath = cf.getFileList(strWorkingDir, strFileNamePattern)
# params
lsSCWnd = np.arange(0.05, 0.3, 0.05)
dRectWnd = 2.0
dSMWnd = 2.0
strCoder = ecc.CODER_GOLAY
m = 1
n = 23
k = 12
r = 2
nInterleaving = 25
print "%s: n=%d, k=%d, m=%d, r=%d, interleave=%d" % \
(strCoder, n, k, m, r, nInterleaving)
# test
lsResult = []
for dCodingWnd in lsSCWnd:
print "evalauting SCWnd=%.2f..." % dCodingWnd
for fn in lsFilePath:
lsDataResult = sd.evaluateSingleData(strWorkingDir, fn,
dRectDuration = dRectWnd,
dSMDurction = dSMWnd,
dSCDuration = dCodingWnd,
eng=eng, strCoder=strCoder,
n=n, k=k, m=m, r=r,
nInterleaving=nInterleaving)
lsResult.extend(lsDataResult)
dfResult = pd.DataFrame(lsResult)
gp = dfResult.groupby(dfResult[sd.WND_SC])
dfMean = gp.mean()
return dfMean, dfResult
def validateOnSH(strInPath, strOutPath, bSerialize=False):
'''
this function validates user diversity on Shanghai data set
param:
strInPath - path for separate files of top users
strOutPath - path to serialize model
Note: we need another script to distributes records of top users into separate files,
and this function will only read separate files from strInPath
'''
# find xdr
lsXDR = common_function.getFileList(strInPath, "out")
dcVariableImportance = {} # variable importance of each personal model
dcModels = {} # dict of personal models
for xdr in lsXDR:
# load data
print("processing %s..." % xdr)
dfData = pd.read_csv(xdr, sep='|', \
names= ['BEGIN_TIME','BEGIN_TIME_MSEL','MSISDN','IMSI','SERVER_IP',\
'SERVER_PORT','APN','PROT_CATEGORY','PROT_TYPE','LAC','SAC',\
'CI','IMEI','RAT','HOST','STREAMING_URL','STREAMING_FILESIZE',\
'STREAMING_DW_PACKETS','STREAMING_DOWNLOAD_DELAY','ASSOCIATED_ID',\
'L4_UL_THROUGHPUT','L4_DW_THROUGHPUT', 'use_less'] )
del dfData['use_less']
dfData['DOWNLOAD_RATIO'] = dfData.iloc[:,17]*1.0/dfData.iloc[:,16]
strIMSI = xdr.split('/')[-1].split('.')[0]
# prepare data set
mtX, arrY, lsTrainingFeatureNames = preprocessDataSet(dfData, g_lsSelectedColumns, \
g_lsNumericColumns, \
g_lsCategoricalColumns,\
'DOWNLOAD_RATIO')
# # train model
# model = trainModel(mtX, arrY, g_modelParams)
# dcVariableImportance[strIMSI] = getVariableImportance(model, lsTrainingFeatureNames)
#
# # test
# mse = mean_squared_error(arrY, model.predict(mtX) )
# mae = mean_absolute_error(arrY, model.predict(mtX) )
# print("MSE: %.4f, MAE: %.4f" % (mse, mae) )
# cross validation
dcPersonalModels = crossValidate(mtX, arrY, g_modelParams, 10)
bestModel, fBestScore = max(dcPersonalModels.iteritems(), key=operator.itemgetter(1) )
dcVariableImportance[strIMSI] = getVariableImportance(bestModel, lsTrainingFeatureNames)
dcModels[strIMSI] = (fBestScore, bestModel)
print("model:%s, #record=%d, best=%0.2f, mean=%.2f, std=%0.2f. \n)" % \
(strIMSI, len(arrY), fBestScore, np.mean(dcPersonalModels.values() ), np.std(dcPersonalModels.values() ) ) )
dfVariableImportance = pd.DataFrame(dcVariableImportance).T
# serialize models
if(bSerialize is True):
common_function.serialize2File(strOutPath+'serDcModels.out', dcPersonalModels)
dfVariableImportance.to_csv(strOutPath+'dfVariableImportance_all.out')
return dcModels, dfVariableImportance
def evaluateReconciliationParams(eng):
"""
Evaluate the effect of reconciliation parameters, i.e., m, k, on
system performance
Paramters:
---------
lsFileList:
list of data files
eng:
instance of matlab engine
strCoder:
name of coder
lsM:
the possible values of m
lsR:
the possible values of r
Returns:
a pandas.DataFrame consisted of all performance result
"""
# select data
strWorkingDir = "../../data/evaluation/reconciliation/"
strFileNamePattern= None
lsFilePath = cf.getFileList(strWorkingDir, strFileNamePattern)
# parameter
strCoder = ecc.CODER_RS
lsM = [4,]
lsR = range(1, 8)
dRectWnd = 2.0
dSMWnd = 2.0
dSCWnd = 0.15
# evaluate
lsResult = []
if (strCoder == ecc.CODER_RS):
for m in lsM:
for r in lsR:
n = 2**m - 1
k = n - 2*r
if(k<1 or n*m>=500):
break
print "testing m=%d, r=%d..." % (m, r)
for fn in lsFilePath:
lsDataResult = sd.evaluateSingleData(strWorkingDir, fn,
dRectDuration=dRectWnd, dSMDuration=dSMWnd,
dSCDuration=dSCWnd,
eng=eng, strCoder=strCoder, n=n, k=k, m=m, r=r)
lsResult.extend(lsDataResult)
elif strCoder == ecc.CODER_GOLAY:
n = 23
k = 12
m = 1
r = 2
for fn in lsFilePath:
lsDataResult = sd.evaluateSingleData(strWorkingDir, fn,
dRectDuration=dRectWnd, dSMDuration=dSMWnd,
dSCDuration=dSCWnd,
eng=eng, strCoder=strCoder, n=n, k=k, m=m, r=r)
lsResult.extend(lsDataResult)
# result
dfResult = pd.DataFrame(lsResult)
dcMatchingRate = {}
for r in lsR:
nMatchedKey = (dfResult[sd.ERR_USER_EC][ (dfResult[sd.R]==r) & \
(dfResult[sd.ERR_USER_EC]==0) ]).count()
nTotalKey = dfResult[sd.ERR_USER_EC][dfResult[sd.R]==r].count()
dMatchingRate = nMatchedKey * 1.0 / nTotalKey
dcMatchingRate[r] = dMatchingRate
srMatchingRate = pd.Series(dcMatchingRate)
return dfSummary, dfResult, srMatchingRate