def __mkEdge(key, ef1, ef2, el, ec, ed, ev, ei, norm, mapFile, oFile):
# mcal cat用のlabel項目の作成
label = []
if el:
for nml in el:
label.append("$s{" + nml + "}")
evcdStr = []
#エッジの拡大率は固定
er = 10
if ev:
evcdStr.append(ev + ":ev")
if ec:
evcdStr.append(ec + ":ec")
if ed:
evcdStr.append(ed + ":ed")
f = None
if el:
f <<= nm.mcal(c='cat(\"_\",%s)' % (','.join(label)), a="##label", i=ei)
else:
f <<= nm.msetstr(v="", a="##label", i=ei)
if len(evcdStr) == 0:
f <<= nm.mcut(f="%s:key,%s:nam1,%s:nam2,##label:el" % (key, ef1, ef2))
else:
f <<= nm.mcut(f="%s:key,%s:nam1,%s:nam2,##label:el,%s" %
(key, ef1, ef2, ','.join(evcdStr)))
if not ev:
f <<= nm.msetstr(v="", a="ev")
if not ed:
f <<= nm.msetstr(v="", a="ed")
if not ec:
f <<= nm.msetstr(v="", a="ec")
f <<= nm.mnullto(f="key", v="##NULL##")
f <<= nm.mjoin(k="key", K="nam", m=mapFile, f="num:keyNum")
f <<= nm.mjoin(k="nam1", K="nam", m=mapFile, f="num:num1,leaf:leaf1")
f <<= nm.mjoin(k="nam2", K="nam", m=mapFile, f="num:num2,leaf:leaf2")
if norm:
f <<= nm.mnormalize(f="ev:ev2", c="range")
f <<= nm.mcal(c='${ev2}*(%s-1)+1' % (er), a="evv")
f <<= nm.mcut(
f="key,nam1,nam2,keyNum,num1,num2,el,evv:ev,ed,ec,leaf1,leaf2",
o=oFile)
else:
f <<= nm.mcut(
f="key,nam1,nam2,keyNum,num1,num2,el,ev,ed,ec,leaf1,leaf2",
o=oFile)
f.run()
def __mkMap(key, nf, ni, ef1, ef2, ei, oFile):
# leaf nodeの構築
infL = [
nm.mcommon(k=ef1, K=key, m=ei, r=True, i=ei).mcut(f="%s:nam" % (ef1)),
nm.mcommon(k=ef2, K=key, m=ei, r=True, i=ei).mcut(f="%s:nam" % (ef2))
]
if ni:
infL.append(
nm.mcommon(k=nf, K=key, m=ei, r=True,
i=ni).mcut(f="%s:nam" % (nf)))
xleaf = nm.muniq(i=infL, k="nam")
xleaf <<= nm.msetstr(v=1, a="leaf")
if ni:
inp = [
nm.mcut(f="%s:nam" % (nf), i=ni),
nm.mcut(f="%s:nam" % (key), i=ni)
]
else:
inp = [
nm.mcut(f="%s:nam" % (ef1), i=ei),
nm.mcut(f="%s:nam" % (ef2), i=ei),
nm.mcut(f="%s:nam" % (key), i=ei)
]
f = None
f <<= nm.muniq(k="nam", i=inp)
f <<= nm.mjoin(k="nam", m=xleaf, f="leaf", n=True)
# nullは最初に来るはずなので、mcalでなくmnumberでもnullを0に採番できるはずだが念のために
f <<= nm.mcal(c='if(isnull($s{nam}),0,line()+1)', a="num")
f <<= nm.mnullto(f="nam", v="##NULL##", o=oFile)
f.run()
def __convertToNumeric(self):
wf1 = self.__tempW.file()
wf2 = self.__tempW.file()
wf3 = self.__tempW.file()
self.mFile = self.__temp.file()
nm.mcut(f="e1:node", i=self.__wfE, o=wf1).run()
nm.mcut(f="e2:node", i=self.__wfE, o=wf2).run()
mcmd = None
if self.__wfN is not None:
nm.mcut(f="n:node", i=self.__wfN, o=wf3).run()
mcmd <<= nm.mcat(i=wf1 + "," + wf2 + "," + wf3, f="node")
else:
mcmd <<= nm.mcat(i=wf1 + "," + wf2, f="node")
mcmd <<= nm.muniq(k="node")
mcmd <<= nm.mnumber(q=True, a="id", o=self.mFile)
mcmd.run()
#エッジファイル変換・保存
self.eFile = self.__temp.file()
mcmd = nm.mjoin(i=self.__wfE,
m=self.mFile,
f="id:id1",
k="e1",
K="node")
mcmd <<= nm.mjoin(m=self.mFile, f="id:id2", k="e2", K="node")
mcmd <<= nm.muniq(k="id1,id2")
mcmd <<= nm.msortf(f="id1%n,id2%n")
mcmd <<= nm.mcut(f="id1:e1,id2:e2,no:row_index", o=self.eFile)
mcmd.run()
#take内部用エッジファイル保存
self.eFileT = self.__temp.file()
mcmd = nm.mcal(i=self.eFile, c="cat(\" \",$s{e1},$s{e2})", a="edge")
mcmd <<= nm.mcut(nfno=True, f="edge", o=self.eFileT)
mcmd.run()
#ノードファイル変換・保存
self.nFile = self.__temp.file()
if self.__wfN is not None:
mcmd = nm.mjoin(i=self.__wfN,
m=self.mFile,
f="id",
k="n",
K="node")
mcmd <<= nm.muniq(k="id")
mcmd <<= nm.msortf(f="id%n")
mcmd <<= nm.mcut(f="id:n,no:row_index", o=self.nFile)
mcmd.run()
else:
#エッジよりノード生成
wf4 = self.__tempW.file()
wf5 = self.__tempW.file()
nm.mcut(i=self.eFile, f="e1:n", o=wf4).run()
nm.mcut(i=self.eFile, f="e2:n", o=wf5).run()
mcmd = nm.mcat(i="%s,%s" % (wf4, wf5))
mcmd <<= nm.muniq(k="n")
mcmd <<= nm.msortf(f="n%n")
mcmd <<= nm.mnumber(q=True, a="row_index", o=self.nFile)
mcmd.run()
#ワークファイル削除
self.__tempW.rm()
def mkIndex(oFile, iFile):
f = None
f <<= nm.mcat(i=iFile)
f <<= nm.mcut(f="date,c")
f <<= nm.mavg(k="date", f="c")
f <<= nm.mcal(c="round(${c},1)", a="i")
f <<= nm.mcut(f="date,i", o="%s" % oFile)
f.run()
def calTime(iFile, oFile):
f = None
f <<= nm.mnumber(q=True, a="id", i=iFile)
f <<= nm.mcal(c='$s{mean}+"("+$s{sd}+")"', a="time")
f <<= nm.m2cross(k="method", s="dataSize", f="time")
f <<= nm.msortf(f="id%n")
f <<= nm.mcut(f="method,10000:small,1000000:middle,100000000:large")
f <<= nm.mfldname(q=True, o=oFile)
f.run()
def convSim(self, ifile, ofile, map1, logDir):
f = None
f <<= nm.mcut(nfni=True, f="0:tra", i=ifile)
f <<= nm.msed(f="tra", c=' $', v="")
f <<= nm.mnumber(q=True, S=1, a="num1")
f <<= nm.mtra(r=True, f="tra:num11")
f <<= nm.mnumber(q=True, S=1, a="order")
f <<= nm.mcal(c='${num11}+1', a="num1")
f <<= nm.mjoin(k="num1", m=map1, f=self.ef1)
f <<= nm.mtra(k="num0", s="order%n,num1%n", f=self.ef1)
f <<= nm.mcut(f=self.ef1, o="{}/{}".format(logDir, ofile))
f.run()
def calRelative(iFile, oFile):
mcut = None
mcut <<= nm.mselstr(f="method", v="mcut", i="methods.csv")
f = None
f <<= nm.mnumber(q=True, a="id", i=iFile)
f <<= nm.mjoin(k="dataSize", m=mcut, f="mean:base")
f <<= nm.mcal(c='round(${mean}/${base},0.1)', a="score")
f <<= nm.m2cross(k="method", s="dataSize", f="score")
f <<= nm.msortf(f="id%n")
f <<= nm.mcut(f="method,10000:small,1000000:middle,100000000:large")
f <<= nm.mfldname(q=True, o=oFile)
f.run()
def mkdata(rowSize, oFile):
f = None
f <<= nm.mnewnumber(a="id", l=rowSize)
f <<= nm.mrand(min=0, max=9, int=True, S=1111, a="key1")
f <<= nm.mrand(min=100, max=999, int=True, S=1113, a="key2")
f <<= nm.mrand(min=10000, max=99999, int=True, S=1117, a="key3")
f <<= nm.mcal(c='randi(0,999,101)', a="int1")
f <<= nm.mcal(c='randi(-999,999,111)', a="int2")
f <<= nm.mcal(c='rand(991)', a="float1")
f <<= nm.mcal(c='rand(997)*200-100', a="float2")
f <<= nm.mcal(c='0d20100101+randi(0,3650,137)', a="date")
f <<= nm.mcal(c='right(t2s(0t000000+randi(0,86399,133)),6)',
a="time",
o=oFile)
f.run()
def enumerate(self, eArgs):
pFiles = []
tFiles = []
tf = mtemp.Mtemp()
for cName, posSize in self.db.clsNameRecSize.items():
negSize = self.db.traSize - posSize
if "minGR" in eArgs:
self.minGR = eArgs["minGR"]
else:
minProb = eArgs["minProb"] if ("minProb" in eArgs) else 0.5
if "uniform" in eArgs and eArgs["uniform"] == True:
self.minGR = (minProb / (1 - minProb)) * (
self.db.clsSize - 1) # マニュアルの式(4)
else:
self.minGR = (minProb / (1 - minProb)) * (
float(negSize) / float(posSize)) # マニュアルの式(4)
# 最小サポートと最小サポート件数
# s=0.05
# s=c1:0.05,c2:0.06
# S=10
# S=c1:10,c2:15
if "minCnt" in eArgs:
if isinstance(eArgs["minCnt"], dict):
self.minPos = eArgs["minCnt"][cName]
else:
self.minPos = eArgs["minCnt"]
else:
if isinstance(eArgs["minSup"], dict):
self.minPos = int(eArgs["minSup"][cName] * float(posSize) +
0.99)
else:
self.minPos = int(eArgs["minSup"] * flost(posSize) + 0.99)
# 最大サポートと最大サポート件数
if "maxCnt" in eArgs:
if isinstance(eArgs["maxCnt"], dict):
self.maxPos = eArgs["maxCnt"][cName]
else:
self.maxPos = eArgs["maxCnt"]
elif "maxSup" in eArgs:
if isinstance(eArgs["maxSup"], dict):
self.maxPos = int(eArgs["maxSup"][cName] * float(posSize) +
0.99)
else:
self.maxPos = int(eArgs["maxSup"] * float(posSize) + 0.99)
else:
self.maxPos = None
self.sigma[cName] = self.calSigma(self.minPos, self.minGR, posSize,
negSize)
# lcmのパラメータ設定と実行
# 頻出パターンがなかった場合、lcm出力ファイルが生成されないので
# そのときのために空ファイルを生成しておいく。
lcmout = tf.file() # lcm出力ファイル
with open(lcmout, "w") as efile:
pass
runPara = {}
if self.msgoff:
runPara["type"] = eArgs["type"] + "IA_"
else:
runPara["type"] = eArgs["type"] + "IA"
#if self.maxPos: #rubyだとif @maxCntなってる（どこにも設定されてないので）動いてないはず
if self.maxPos:
runPara["U"] = self.maxPos
if "minLen" in eArgs:
runPara["l"] = str(eArgs["minLen"])
if "maxLen" in eArgs:
runPara["u"] = str(eArgs["maxLen"])
runPara["w"] = self.weightFile[cName]
runPara["i"] = self.file
runPara["sup"] = str(self.sigma[cName])
runPara["o"] = lcmout
# lcm実行
#MCMD::msgLog("#{run}")
#TAKE::run_lcm(run)
#print(self.sigma)
#print(runPara)
#MCMD::msgLog("output patterns to CSV file ...")
extTake.lcm(runPara)
pFiles.append(self.temp.file())
transle = tf.file()
extTake.lcmtrans(lcmout, "e", transle)
f = nm.mdelnull(f="pattern", i=transle)
f <<= nm.mcal(c='round(${countN},1)', a="neg")
f <<= nm.mcal(c='round(${countP}/%s,1)' % (self.posWeight[cName]),
a="pos")
f <<= nm.mdelnull(f="pattern") #いる？
f <<= nm.msetstr(v=cName, a="class")
f <<= nm.msetstr(v=posSize, a="posTotal")
f <<= nm.msetstr(v=self.minGR, a="minGR")
f <<= nm.mcut(f="class,pid,pattern,size,pos,neg,posTotal,minGR",
o=pFiles[-1])
f.run()
#s = nutil.mrecount(i=self.file)
#MCMD::msgLog("the number of contrast patterns on class `#{cName}' enumerated is #{s}")
if self.outtf:
# トランザクション毎に出現するパターンを書き出す
#MCMD::msgLog("output tid-patterns ...")
tFiles.append(self.temp.file())
xxw = tf.file()
xxw = nm.mcut(f=self.db.idFN, i=self.db.file)
xxw <<= nm.muniq(k=self.db.idFN)
xxw <<= nm.mnumber(S=0, a="__tid", q=True)
translt = self.temp.file()
extTake.lcmtrans(lcmout, "t", translt)
f = nm.mjoin(k="__tid", m=xxw, f=self.db.idFN, i=translt)
f <<= nm.msetstr(v=cName, a="class")
f <<= nm.mcut(f=self.db.idFN + ",class,pid", o=tFiles[-1])
f.run()
# クラス別のパターンとtid-pidファイルを統合して最終出力
self.pFile = self.temp.file()
self.tFile = self.temp.file()
# パターンファイル併合
xxpCat = tf.file()
f = nm.mcat(i=",".join(pFiles))
f <<= nm.msortf(f="class,pid")
f <<= nm.mnumber(s="class,pid", S=0, a="ppid", o=xxpCat)
f.run()
# パターンファイル計算
items = self.db.items
f = nm.mcut(f="class,ppid:pid,pattern,size,pos,neg,posTotal,minGR",
i=xxpCat)
f <<= nm.msetstr(v=self.db.traSize, a="total")
f <<= nm.mcal(c='${total}-${posTotal}', a="negTotal") # negのトータル件数
f <<= nm.mcal(c='${pos}/${posTotal}', a="support") # サポートの計算
f <<= nm.mcal(
c=
'if(${neg}==0,1.797693135e+308,(${pos}/${posTotal})/(${neg}/${negTotal}))',
a="growthRate")
if "uniform" in eArgs and eArgs["uniform"] == True:
f <<= nm.mcal(
c='(${pos}/${posTotal})/(${pos}/${posTotal}+(%s-1)*${neg}/${negTotal})'
% (self.db.clsSize),
a="postProb")
else:
f <<= nm.mcal(c='${pos}/(${pos}+${neg})', a="postProb")
f <<= nm.msel(c='${pos}>=%s&&${growthRate}>=${minGR}' %
(self.minPos)) # minSupとminGRによる選択
f <<= nm.mvreplace(vf="pattern",
m=items.file,
K=items.idFN,
f=items.itemFN)
f <<= nm.mcut(
f="class,pid,pattern,size,pos,neg,posTotal,negTotal,total,support,growthRate,postProb"
)
f <<= nm.mvsort(vf="pattern")
f <<= nm.msortf(f="class%nr,postProb%nr,pos%nr", o=self.pFile)
f.run()
# アイテムを包含している冗長なタクソノミを削除
if items.taxonomy:
#MCMD::msgLog("reducing redundant rules in terms of taxonomy ...")
##ここは後で
zdd = VSOP.constant(0)
dt = nm.mcut(i=self.pFile, f="pattern")
for fldVal in dt:
zdd = zdd + VSOP.itemset(fldVal[0])
zdd = self.reduceTaxo(zdd, self.db.items)
xxp1 = tf.file()
xxp2 = tf.file()
xxp3 = tf.file()
zdd.csvout(xxp1)
nm.mcut(nfni=True, f="1:pattern",
i=xxp1).mvsort(vf="pattern").msortf(f="pattern",
o=xxp2).run()
nm.msortf(f="pattern", i=self.pFile).mcommon(
k="pattern", m=xxp2).msortf(f="class%nr,postProb%nr,pos%nr",
o=xxp3).run()
shutil.move(xxp3, self.pFile)
if self.outtf:
# 列挙されたパターンを含むtraのみ選択するためのマスタ
xxp4 = nm.mcut(f="class,pid", i=self.pFile)
f = nm.mcat(i=",".join(tFiles))
f <<= nm.mjoin(k="class,pid", m=xxpCat,
f="ppid") # 全クラス統一pid(ppid)結合
f <<= nm.mcommon(k="class,ppid", K="class,pid",
m=xxp4) # 列挙されたパターンの選択
f <<= nm.mcut(f=self.db.idFN + ",class,ppid:pid")
f <<= nm.msortf(f=self.db.idFN + ",class,pid", o=self.tFile)
f.run()
def enumerate(self,eArgs):
"""
eArgsで与えられた条件で、頻出アイテム集合の列挙を実行する。
:type eArgs: dict
:type eArgs['type']: str
:type eArgs['minCnt']: int
:type eArgs['minSup']: float
:type eArgs['maxCnt']: int
:type eArgs['maxSup']: float
:type eArgs['minLen']: int
:type eArgs['maxLen']: int
:type eArgs['top']: int
:type eArgs['skipTP']: bool【default:False】
:param eArgs: 各種列挙パラメータ
:param eArgs['type']: 抽出するアイテム集合の型【'F':頻出集合, 'C':飽和集合, 'M':極大集合】
:param eArgs['minCnt']: 最小サポート(件数)
:param eArgs['minSup']: 最小サポート(確率)
:param eArgs['maxCnt']: 最大サポート(件数)
:param eArgs['maxSup']: 最大サポート(確率)
:param eArgs['minLen']: アイテム集合の最小アイテム数(件数)
:param eArgs['maxLen']: アイテム集合の最大アイテム数(件数)
:param eArgs['top']: 列挙するサポート上位件数(件数)
:param eArgs['skipTP']: トランザクションにマッチするパターン(アイテム集合)の出力を行わない。
"""
tf=mtemp.Mtemp()
self.eArgs=eArgs
self.type = eArgs["type"]
if "minCnt" in eArgs and eArgs["minCnt"] != None:
self.minCnt = int(eArgs["minCnt"])
self.minSup = float(self.minCnt) / float(self.db.traSize)
else:
self.minSup = float(eArgs["minSup"])
self.minCnt = int(self.minSup * float(self.db.traSize) + 0.99)
# 最大サポートと最大サポート件数
self.maxCnt=None
if ("maxCnt" in eArgs and eArgs["maxCnt"]!= None) or ( "maxSup" in eArgs and eArgs["maxSup"]!= None):
if "maxCnt" in eArgs and eArgs["maxCnt"]!= None:
self.maxCnt = int(eArgs["maxCnt"])
self.maxSup = float(self.maxCnt) / float(self.db.traSize)
else:
self.maxSup = float(eArgs["maxSup"])
self.maxCnt = int(self.maxSup * float(self.db.traSize) + 0.99)
params = {}
if self.msgoff:
params["type"] ="%sIf_"%(self.type)
else:
params["type"] ="%sIf"%(self.type)
if self.maxCnt :
params["U"] = str(self.maxCnt)
if "minLen" in eArgs and eArgs["minLen"] != None :
params["l"] = str(eArgs['minLen'])
if "maxLen" in eArgs and eArgs["maxLen"] != None :
params["u"] = str(eArgs['maxLen'])
# 列挙パターン数上限が指定されれば、一度lcmを実行して最小サポートを得る
if "top" in eArgs and eArgs["top"] != None :
self.top = eArgs["top"]
if self.top and self.top>0 :
xxtop = tf.file()
import copy
top_params = copy.deepcopy(params)
top_params["i"] = self.file
top_params["sup"] = "1"
top_params["K"] = str(self.top)
top_params["so"] = xxtop
import re
top_params["type"] = re.sub('_$', '', top_params["type"] )
extTake.lcm(top_params)
with open(xxtop, "r") as rfile:
self.minCnt = int(rfile.read().strip())
if self.minCnt<0 :
self.minCnt=1
self.skipTP=False
if "skipTP" in eArgs:
self.skipTP=eArgs["skipTP"]
# lcm_seq出力ファイル
lcmout = tf.file()
# 頻出パターンがなかった場合、lcm出力ファイルが生成されないので
# そのときのために空ファイルを生成しておいく。
with open(lcmout, "w") as efile:
pass
# lcm実行
params["i"] = self.file
params["sup"] = str(self.minCnt)
params["o"] = lcmout
extTake.lcm(params)
# caliculate one itemset for lift value
xxone= tf.file()
tpstr = "FIf_" if self.msgoff else "FIf"
extTake.lcm(type=tpstr,i=self.file,sup=1,o=xxone,l=1,u=1)
# パターンのサポートを計算しCSV出力する
#MCMD::msgLog("output patterns to CSV file ...")
xxp0 = tf.file()
self.pFile = self.temp.file()
items=self.db.items
trans0 = self.temp.file()
extTake.lcmtrans(lcmout,"p",trans0)
f = nm.mdelnull(i=trans0,f="pattern")
f <<= nm.mvreplace(vf="pattern",m=items.file,K=items.idFN,f=items.itemFN)
f <<= nm.msetstr(v=self.db.traSize,a="total")
f <<= nm.mcal(c='${count}/${total}',a="support")
f <<= nm.mcut(f="pid,pattern,size,count,total,support")
f <<= nm.mvsort(vf="pattern")
f <<= nm.msortf(f="pid",o=xxp0)
f.run()
# xxp0
# pid,count,total,support,pattern
# 0,13,13,1,A
# 4,6,13,0.4615384615,A B
xxp1=tf.file()
# taxonomy指定がない場合(2010/11/20追加)
if items.taxonomy==None:
shutil.move(xxp0,xxp1)
# taxonomy指定がある場合
else:
#MCMD::msgLog("reducing redundant rules in terms of taxonomy ...")
zdd=VSOP.constant(0)
fobj = nm.mcut(i=xxp0,f='pattern')
for fldVal in fobj:
zdd=zdd+VSOP.itemset(fldVal[0])
zdd=self.reduceTaxo(zdd,self.db.items)
xxz1=tf.file()
xxz2=tf.file()
zdd.csvout(xxz1)
f0=None
f0 <<= nm.mcut(nfni=True,f="1:pattern",i=xxz1)
f0 <<= nm.mvsort(vf="pattern")
f0 <<= nm.msortf(f="pattern")
f=None
f <<= nm.msortf(f="pattern",i=xxp0)
f <<= nm.mcommon(k="pattern",m=f0)
f <<= nm.msortf(f="pid",o=xxp1)
f.run()
# lift値の計算
transl = tf.file()
extTake.lcmtrans(xxone,"p",transl)
xxp2 = nm.mdelnull(i=transl,f="pattern")
xxp2 <<= nm.mvreplace(vf="pattern",m=items.file,K=items.idFN,f=items.itemFN)
xxp2 <<= nm.msortf(f="pattern")
xxp3 = nm.mcut(f="pid,pattern",i=xxp1)
xxp3 <<= nm.mtra(f="pattern",r=True)
xxp3 <<= nm.mjoin(k="pattern",m=xxp2,f="count:c1")
xxp3 <<= nm.mcal(c='ln(${c1})',a="c1ln")
xxp3 <<= nm.msum(k="pid",f="c1ln")
# p3
# pid,pattern,c1,c1ln
# 0,A,13,2.564949357
# 1,E,7,1.945910149
#おかしくなる?=>OK
f3 = nm.mjoin(k="pid",f="c1ln",i=xxp1,m=xxp3)
f3 <<= nm.mcal(c='round(exp(ln(${count})-${c1ln}+(${size}-1)*ln(${total})),0.0001)',a="lift")
f3 <<= nm.mcut(f="pid,size,count,total,support,lift,pattern")
f3 <<= nm.msortf(f="support%nr",o=self.pFile)
f3.run()
#self.size = mrecount.mrecount(i=self.file)
#MCMD::msgLog("the number of patterns enumerated is #{@size}")
if not self.skipTP:
# トランザクション毎に出現するシーケンスを書き出す
#MCMD::msgLog("output tid-patterns ...")
self.tFile = self.temp.file()
xxw3i = tf.file()
extTake.lcmtrans(lcmout,"t",xxw3i)
xxw1 = nm.mcut(f=self.db.idFN,i=self.db.file).muniq(k=self.db.idFN).mnumber(S=0,a="__tid",q=True).msortf(f="__tid")
xxw2 = nm.mcut(f="pid",i=self.pFile)
xxw3 = nm.mcommon(k="pid",i=xxw3i,m=xxw2).mjoin(k="__tid",m=xxw1,f=self.db.idFN).mcut(f=self.db.idFN+",pid",o=self.tFile)
xxw3.run()
def run(self):
from datetime import datetime
t = datetime.now()
temp=nu.Mtemp()
xxsspcin=temp.file()
xxmap=temp.file()
# traファイルの変換とマップファイルの作成
if self.num :
total = self.convN(self.iFile,self.idFN,self.itemFN,xxsspcin,xxmap)
else:
total = self.conv(self.iFile,self.idFN,self.itemFN,xxsspcin,xxmap)
# system "head xxsspcin"
# 3 5 0 2
# 4 1 2
# 0 2 3 1
# 1 0 2
# 3 4 0 1
# system "head xxmap"
# ##item,##freq%0nr,##num
# b,4,0
# d,4,1
# f,4,2
minSupp = int(total*self.minSupPrb) if self.minSupPrb else self.minSupCnt
# sspc用simの文字列
if self.sim :
if self.sim=="J":
sspcSim="R"
elif self.sim=="P":
sspcSim="P"
elif self.sim=="C":
sspcSim="i"
# sim=省略時はRでth=0とする(sim制約なし)
else:
sspcSim="R"
self.th=0
############ 列挙本体 ############
xxsspcout=temp.file()
tpstr = sspcSim+"ft_" if self.msgoff else sspcSim+"ft"
extTake.sspc(type=tpstr,TT=minSupp,i=xxsspcin,th=self.th,o=xxsspcout)
##################################
xxtmmp=temp.file()
f = nm.mread(i=xxsspcout)
f <<= nm.cmd("tr ' ()' ','")
f <<= nm.mcut(f="1:i1,2:i2,0:frequency,4:sim",nfni=True)
if self.num :
f <<= nm.mfldname(f="i1:node1,i2:node2")
if self.sim!="C":
f <<= nm.mfsort(f="node1,node2")
f <<= nm.mjoin(k="node1",K="##item",m=xxmap,f="##freq:frequency1")
f <<= nm.mjoin(k="node2",K="##item",m=xxmap,f="##freq:frequency2")
else:
f <<= nm.mjoin(k="i1",K="##num",m=xxmap,f="##item:node1,##freq:frequency1")
f <<= nm.mjoin(k="i2",K="##num",m=xxmap,f="##item:node2,##freq:frequency2")
if self.sim!="C":
f <<= nm.mcut(f="i1,i2,frequency,sim,node1,node2,frequency1,frequency2,node1:node1x,node2:node2x")
f <<= nm.mfsort(f="node1x,node2x")
f <<= nm.mcal(c='if($s{node1}==$s{node1x},$s{frequency1},$s{frequency2})',a="freq1")
f <<= nm.mcal(c='if($s{node2}==$s{node2x},$s{frequency2},$s{frequency1})',a="freq2")
f <<= nm.mcut(f="i1,i2,frequency,sim,node1x:node1,node2x:node2,freq1:frequency1,freq2:frequency2")
f <<= nm.msetstr(v=total,a="total")
f <<= nm.mcal(c='${frequency}/${frequency1}',a="confidence")
f <<= nm.mcal(c='${frequency}/${total}',a="support")
f <<= nm.mcal(c='${frequency}/(${frequency1}+${frequency2}-${frequency})',a="jaccard")
f <<= nm.mcal(c='(${frequency}*${total})/((${frequency1}*${frequency2}))',a="lift")
f <<= nm.mcal(c='(ln(${frequency})+ln(${total})-ln(${frequency1})-ln(${frequency2}))/(ln(${total})-ln(${frequency}))',a="PMI")
f <<= nm.mcut(f="node1,node2,frequency,frequency1,frequency2,total,support,confidence,lift,jaccard,PMI")
f <<= nm.msortf(f="node1,node2",o=self.oeFile)
f.run()
if self.onFile:
f4 = nm.mcut(f=self.itemFN+":node",i=self.iFile)
f4 <<= nm.mcount(k="node",a="frequency")
if self.node_support :
minstr = "[%s,]"%(minSupp)
f4 <<= nm.mselnum(f="frequency",c=minstr)
f4 <<= nm.msetstr(v=total,a="total")
f4 <<= nm.mcal(c='${frequency}/${total}',a="support")
f4 <<= nm.mcut(f="node,support,frequency,total",o=self.onFile)
f4.run()
procTime=datetime.now()-t
# ログファイル出力
if self.logFile :
kv=[["key","value"]]
for k,v in self.args.items():
kv.append([k,str(v)])
kv.append(["time",str(procTime)])
nm.writecsv(i=kv,o=self.logFile).run()
os.environ["KG_VerboseLevel"] = "3"
debug = "on"
iPath = "./DATA"
oPath = "./OUTPUT/apriori"
os.system("mkdir -p %s" % oPath)
iFile = "%s/price_large.csv" % (iPath)
topix = "%s/index.csv" % (iPath)
# make a transaction data, which date as a transaction and tickerID as an item
tra = None
tra <<= nm.mcut(f="id,date,c", i=iFile)
tra <<= nm.mjoin(k="date", m=topix, f="i")
tra <<= nm.mslide(k="id", s="date", f="date:date2,c:c2,i:i2")
tra <<= nm.mcal(c="${c2}/${c}-${i2}/${i}", a="ret")
tra <<= nm.mselnum(f="ret", c="[0.05,0.1]")
tra <<= nm.mcut(f="id,date2:date,ret")
# frequency of one item
freq = None
freq <<= nm.mcut(f="id", i=tra)
freq <<= nm.mcount(k="id", a="freq")
freq <<= nm.mselnum(f="freq", c="[5,]")
# total number of transactions
total = None
total <<= nm.mcut(f="date", i=tra)
total <<= nm.muniq(k="date")
total <<= nm.mcount(a="total")
def enumerate(self,eArgs):
tf=nu.Mtemp()
# 最小サポートと最小サポート件数
if "minCnt" in eArgs :
self.minCnt = int(eArgs["minCnt"])
self.minSup = float(self.minCnt)/ float(self.db.size)
else:
self.minSup = float(eArgs["minSup"])
self.minCnt = int(self.minSup * float(self.db.size) + 0.99)
# 最大サポートと最大サポート件数
self.maxCnt=None
if "maxCnt" in eArgs or "maxSup" in eArgs:
if "maxCnt" in eArgs:
self.maxCnt = int(eArgs["maxCnt"])
self.maxSup = float(self.maxCnt)/float(self.db.size)
else:
self.maxSup = float(eArgs["maxSup"])
self.maxCnt = int(self.maxSup * float(self.db.size) + 0.99)
#未使用
#@minProb = eArgs["minProb"].to_f # 事後確率
#@minGR = @minProb/(1-@minProb) # 増加率
#@minGR = eArgs["minGR"].to_f if eArgs["minGR"]
# あるクラスをpos、他のクラスをnegにして、パターン列挙した結果ファイル名を格納する
pFiles=[]
tFiles=[]
for cName,posSize in self.db.clsNameRecSize.items():
negSize=self.db.size-posSize
# minGRの計算
if "minGR" in eArgs:
self.minGR=eArgs["minGR"]
else:
minProb = eArgs["minProb"] if ( "minProb" in eArgs ) else 0.5
if "uniform" in eArgs and eArgs["uniform"]:
self.minGR = (minProb/(1-minProb)) * (self.db.clsSize-1) # マニュアルの式(4)
else:
self.minGR = (minProb/(1-minProb)) * (float(negSize)/float(posSize)) # マニュアルの式(4)
# 最小サポートと最小サポート件数
if "minCnt" in eArgs:
self.minPos = eArgs["minCnt"]
else:
self.minPos = int(eArgs["minSup"] * float(posSize) + 0.99)
# 最大サポートと最大サポート件数
if "maxCnt" in eArgs or "maxSup" in eArgs:
if "maxCnt" in eArgs:
self.maxCnt = int(eArgs["maxCnt"])
else:
self.maxCnt = int(eArgs["maxSup"] * float(posSize) + 0.99)
self.sigma[cName] = self.calSigma(self.minPos,self.minGR,posSize,negSize)
# lcm_seqのパラメータ設定と実行
lcmout = tf.file() # lcm_seq出力ファイル
# 頻出パターンがなかった場合、lcm出力ファイルが生成されないので
# そのときのために空ファイルを生成しておいく。
with open(lcmout, "w") as efile:
pass
params = {}
if self.msgoff:
params["type"] ="CIA_"
else:
params["type"] ="CIA"
if self.maxCnt: # windowサイズ上限
params["U"] = str(self.maxCnt)
if "minLen" in eArgs:
params["l"] = str(eArgs["minLen"])
if 'maxLen' in eArgs:
params["u"] = str(eArgs["maxLen"])
if 'gap' in eArgs:
params["g"] = str(eArgs["gap"])
if 'win' in eArgs:
params["G"] = str(eArgs["win"])
params["w"] = self.weightFile[cName]
params["i"] = self.file
params["sup"] = str(self.sigma[cName])
params["o"] = lcmout
# lcm_seq実行
#MCMD::msgLog("#{run}")
if 'padding' in eArgs and eArgs["padding"]: # padding指定時は、0アイテムを出力しないlcm_seqを実行
extTake.lcmseq_zero(params)
else:
extTake.lcmseq(params)
# パターンのサポートを計算しCSV出力する
#MCMD::msgLog("output patterns to CSV file ...")
pFiles.append(self.temp.file())
transle = self.temp.file()
extTake.lcmtrans(lcmout,"e",transle) # pattern,countP,countN,size,pid
f=None
f <<= nm.mdelnull(f="pattern",i=transle)
f <<= nm.mcal(c='round(${countN},1)',a="neg")
f <<= nm.mcal(c='round(${countP}/%s,1)'%(self.posWeight[cName]),a="pos")
f <<= nm.mdelnull(f="pattern")
f <<= nm.msetstr(v=cName,a="class")
f <<= nm.msetstr(v=posSize,a="posTotal")
f <<= nm.msetstr(v=self.minGR,a="minGR")
f <<= nm.mcut(f="class,pid,pattern,size,pos,neg,posTotal,minGR",o=pFiles[-1])
f.run()
#s = MCMD::mrecount("i=#{pFiles.last}") # 列挙されたパターンの数
#MCMD::msgLog("the number of contrast patterns on class `#{cName}' enumerated is #{s}")
if self.outtf :
# トランザクション毎に出現するシーケンスを書き出す
#MCMD::msgLog("output tid-patterns ...")
tFiles.append(self.temp.file())
xxw= tf.file()
f=None
f <<= nm.mcut(f=self.db.idFN,i=self.db.file)
f <<= nm.muniq(k=self.db.idFN)
f <<= nm.mnumber(S=0,a="__tid",q=True)
f <<= nm.msortf(f="__tid",o=xxw)
f.run()
nm.mcut(f=self.db.idFN,i=self.db.file).muniq(k=self.db.idFN).mnumber(S=0,a="__tid",q=True,o=xxw).run()
translt = self.temp.file()
extTake.lcmtrans(lcmout,"t",translt)
nm.mjoin(k="__tid",m=xxw,f=self.db.idFN,i=translt).msetstr(v=cName,a="class").mcut(f=self.db.idFN+",class,pid",o=tFiles[-1]).run()
# クラス別のパターンとtid-pidファイルを統合して最終出力
self.pFile = self.temp.file()
self.tFile = self.temp.file()
# パターンファイル併合
xxpCat = tf.file()
f = nm.mcat(i=",".join(pFiles))
f <<= nm.msortf(f="class,pid")
f <<= nm.mnumber(s="class,pid",S=0,a="ppid",o=xxpCat)
f.run()
# パターンファイル計算
items=self.db.items
f=""
f = nm.mcut(f="class,ppid:pid,pattern,size,pos,neg,posTotal,minGR",i=xxpCat)
f <<= nm.msetstr(v=self.db.size,a="total")
f <<= nm.mcal(c='${total}-${posTotal}',a="negTotal") # negのトータル件数
f <<= nm.mcal(c='${pos}/${posTotal}',a="support") # サポートの計算
f <<= nm.mcal(c='if(${neg}==0,1.797693135e+308,(${pos}/${posTotal})/(${neg}/${negTotal}))',a="growthRate")
if "uniform" in eArgs and eArgs["uniform"] == True:
f <<= nm.mcal(c='(${pos}/${posTotal})/(${pos}/${posTotal}+(%s-1)*${neg}/${negTotal})'%(self.db.clsSize),a="postProb")
else:
f <<= nm.mcal(c='${pos}/(${pos}+${neg})',a="postProb")
f <<= nm.msel(c='${pos}>=%s&&${growthRate}>=${minGR}'%(self.minPos)) # minSupとminGRによる選択
f <<= nm.mvreplace(vf="pattern",m=items.file,K=items.idFN,f=items.itemFN)
f <<= nm.mcut(f="class,pid,pattern,size,pos,neg,posTotal,negTotal,total,support,growthRate,postProb")
f <<= nm.mvsort(vf="pattern")
f <<= nm.msortf(f="class%nr,postProb%nr,pos%nr",o=self.pFile)
f.run()
if self.outtf :
# 列挙されたパターンを含むtraのみ選択するためのマスタ
xxp4=nm.mcut(f="class,pid",i=self.pFile)
f = nm.mcat(i=",".join(tFiles))
f <<= nm.mjoin(k="class,pid",m=xxpCat,f="ppid") # 全クラス統一pid(ppid)結合
f <<= nm.mcommon(k="class,ppid",K="class,pid",m=xxp4) # 列挙されたパターンの選択
f <<= nm.mcut(f=self.db.idFN+",class,ppid:pid")
f <<= nm.msortf(f=self.db.idFN+",class,pid",o=self.tFile)
f.run()
self.size = nu.mrecount(i=self.pFile)
def __mkNode(key, nf, nl, nv, nc, ni, ef1, ef2, ei, noiso, norm, mapFile,
oFile):
xbyE = None
xbyN = None
# edgeファイルからnode情報を生成
# noiso(孤立node排除)の場合は、edgeにあってnodeにないidを省く必要があるので計算する。
if ni == None or (ni != None and noiso):
inp = [
nm.mcut(f="%s:key,%s:nam,%s:nl" % (key, ef1, ef1), i=ei),
nm.mcut(f="%s:key,%s:nam,%s:nl" % (key, ef2, ef2), i=ei)
]
xbyE <<= nm.mnullto(i=inp, f="key", v="##NULL##")
xbyE <<= nm.muniq(k="key,nam")
xbyE <<= nm.mjoin(k="key", K="nam", m=mapFile, f="num:keyNum")
xbyE <<= nm.mjoin(k="nam", K="nam", m=mapFile, f="num,leaf")
xbyE <<= nm.msetstr(v=",,,,", a="nv,nc,nlKey,nvKey,ncKey")
xbyE <<= nm.mcut(
f="key,nam,keyNum,num,nl,nv,nc,leaf,nvKey,ncKey,nlKey")
# nodeファイルから作成
if ni:
# mcal cat用のlabel項目の作成
label = []
#label項目
if nl:
for nml in nl:
label.append(nml)
else:
label.append("$s{%s}" % (nf))
nvcStr = []
if nv:
nvcStr.append('%s:nv' % (nv))
if nc:
nvcStr.append('%s:nc' % (nc))
"""
# map
# nam,leaf,num
# ##NULL##,,0
# #1_1,,2
# #1_2,,3
# #1_3,,4
# #2_1,,5
# a,1,6
# b,1,7
# c,1,8
"""
f = None
f <<= nm.mcal(c='cat("_",%s)' % (','.join(label)), a="##label", i=ni)
if len(nvcStr) == 0:
f <<= nm.mcut(f='%s:key,%s:nam,##label:nl' % (key, nf))
else:
f <<= nm.mcut(f='%s:key,%s:nam,##label:nl,%s' %
(key, nf, ','.join(nvcStr)))
f <<= nm.mnullto(f="key", v="##NULL##")
if not nv:
f <<= nm.msetstr(v="", a="nv")
if not nc:
f <<= nm.msetstr(v="", a="nc")
f <<= nm.mjoin(k="key", K="nam", m=mapFile, f="num:keyNum")
f <<= nm.mjoin(k="nam", K="nam", m=mapFile, f="num,leaf")
if norm:
#ノードの拡大率は固定
nr = 3
f <<= nm.mnormalize(f="nv:nv2", c="range")
f <<= nm.mcal(c='${nv2}*(%s-1)+1' % (nr), a="nvv")
f <<= nm.mcut(f="key,nam,keyNum,num,nl,nvv:nv,nc,leaf") #o=#{xxa}"
else:
f <<= nm.mcut(f="key,nam,keyNum,num,nl,nv,nc,leaf") #o=#{xxa}"
xbyN <<= nm.mjoin(k="keyNum",
K="num",
m=f,
f="nl:nlk,nv:nvKey,nc:ncKey",
n=True,
i=f)
xbyN <<= nm.mcal(c='if(isnull($s{nlk}),$s{key},$s{nlk})', a='nlKey')
xbyN <<= nm.mcut(f="nlk", r=True)
if ni != None and noiso:
nm.mcommon(k="key,nam", m=xbyE, i=xbyN, o=oFile).run()
elif ni != None:
xbyN.writecsv(o=oFile).run()
else:
xbyE.writecsv(o=oFile).run()
def enumerate(self, eArgs):
tf = mtemp.Mtemp()
# 最小サポートと最小サポート件数
if "minCnt" in eArgs and eArgs["minCnt"] != None:
self.minCnt = int(eArgs["minCnt"])
self.minSup = float(self.minCnt) / float(self.db.size)
else:
self.minSup = float(eArgs["minSup"])
self.minCnt = int(self.minSup * float(self.db.size) + 0.99)
# 最大サポートと最大サポート件数
self.maxCnt = None
if ("maxCnt" in eArgs
and eArgs["maxCnt"] != None) or ("maxSup" in eArgs
and eArgs["maxSup"] != None):
if "maxCnt" in eArgs and eArgs["maxCnt"] != None:
self.maxCnt = int(eArgs["maxCnt"])
self.maxSup = float(self.maxCnt) / float(self.db.size)
else:
self.maxSup = float(eArgs["maxSup"])
self.maxCnt = int(self.maxSup * float(self.db.size) + 0.99)
# 列挙パターン数上限が指定されれば、一度lcmを実行して最小サポートを得る
if "top" in eArgs and eArgs["top"] != None:
self.top = eArgs["top"]
# 列挙パターン数上限が指定されれば、一度lcmを実行して最小サポートを得る
if self.top and self.top > 0:
xxtop = tf.file()
extTake.lcmseq(type="Cf",
K=str(self.top),
i=self.file,
sup="1",
so=xxtop)
with open(xxtop, "r") as rfile:
self.minCnt = int(rfile.read().strip())
# lcm_seq出力ファイル
lcmout = tf.file()
# 頻出パターンがなかった場合、lcm出力ファイルが生成されないので
# そのときのために空ファイルを生成しておいく。
with open(lcmout, "w") as efile:
pass
# lcm_seqのパラメータ設定と実行
params = {}
if self.msgoff:
params["type"] = "CIf_"
else:
params["type"] = "CIf"
if self.maxCnt:
params["U"] = str(self.maxCnt)
if "minLen" in eArgs:
params["l"] = str(eArgs["minLen"])
if 'maxLen' in eArgs:
params["u"] = str(eArgs["maxLen"])
if 'gap' in eArgs:
params["g"] = str(eArgs["gap"])
if 'win' in eArgs:
params["G"] = str(eArgs["win"])
params["i"] = self.file
params["sup"] = str(self.minCnt)
params["o"] = lcmout
# lcm_seq実行
#MCMD::msgLog("#{run}")
if 'padding' in eArgs and eArgs[
"padding"]: # padding指定時は、0アイテムを出力しないlcm_seqを実行
extTake.lcmseq_zero(params)
else:
extTake.lcmseq(params)
# パターンのサポートを計算しCSV出力する
self.pFile = self.temp.file()
items = self.db.items
transl = self.temp.file()
extTake.lcmtrans(lcmout, "p", transl)
f = nm.mdelnull(f="pattern", i=transl)
f <<= nm.mvreplace(vf="pattern",
m=items.file,
K=items.idFN,
f=items.itemFN)
f <<= nm.msetstr(v=self.db.size, a="total")
f <<= nm.mcal(c='${count}/${total}', a="support") # サポートの計算
f <<= nm.mcut(f="pid,pattern,size,count,total,support")
f <<= nm.msortf(f="support%nr", o=self.pFile)
f.run()
if self.outtf:
# トランザクション毎に出現するシーケンスを書き出す
#MCMD::msgLog("output tid-patterns ...")
self.tFile = self.temp.file()
xxw = tf.file() #Mtemp.new.name
f = None
f <<= nm.mcut(f=self.db.idFN, i=self.db.file)
f <<= nm.muniq(k=self.db.idFN)
f <<= nm.mnumber(S=0, a="__tid", q=True)
f <<= nm.msortf(f="__tid", o=xxw)
f.run()
translt = self.temp.file()
extTake.lcmtrans(lcmout, "t", translt)
f = None
f <<= nm.msortf(f="__tid", i=translt)
f <<= nm.mjoin(k="__tid", m=xxw, f=self.db.idFN)
f <<= nm.mcut(f=self.db.idFN + ",pid")
f <<= nm.msortf(f=self.db.idFN + ",pid", o=self.tFile)
f.run()
20161013,-0.003854546465
20161014,-0.01164231144
20161017,-0.001684303294
20161018,-0.002248115897
20161019,-0.003859145998
20161020,-0.006883326367
20161021,-0.01175535999
20161024,-0.01474462455
"""
temo = mtemp.Mtemp()
vv = temo.file()
with open(vv,"w") as wfp:
wfp.write(sampDAT1)
rls=[]
xxx1 = n.mcal(i=vv,c="left($s{date},6)",a="month").msetstr(a="ccc",v="")
xxx2 = xxx1.msum({"k":"date","f":"val"})
xxx3 = xxx1.msum({"k":"date","f":"val"},o=rls)
xxx3.run()
print(rls)
for x in xxx2.keyblock(["month"]):
print(x)
for x in xxx2:
if x[3] == "201608":
print(x)
# Filitering the friend pairs of StockCode in the similarity graph of StockCode.
f = None
f <<= nm.msplit(f="pattern", a="item1,item2", i="%s/patterns.csv" % oPath)
f <<= nm.mcut(f="item1,item2,lift", o="%s/rules.csv" % oPath)
f.run(msg="on")
#os.system("mfriends.rb ef=item1,item2 ei=%s/rules.csv ef=item1,item2 sim=lift rank=5 eo=%s/friends.csv -udout"%(oPath,oPath))
##### 上の行は、以下で動くようにする
nt.mfriends(ef="item1,item2",
ei="%s/rules.csv" % oPath,
sim="lift",
rank=5,
udout=True,
eo="%s/friends.csv" % oPath).run()
# visualization of the graph
f = None
f <<= nm.mcal(c="cat(\" \",$s{item1},$s{item2})",
a="edges",
i="%s/friends.csv" % oPath)
f <<= nm.mcut(f="edges", nfno=True, o="%s/edges.csv" % oPath)
f.run(msg="on")
G = nx.read_edgelist("%s/edges.csv" % oPath)
pos = nx.spring_layout(G)
plt.figure(figsize=(10, 10))
nx.draw(G, pos=pos, node_size=40, iterations=20)
plt.savefig("%s/friends.png" % oPath)
def run(self):
temp = mtemp.Mtemp()
### mtra2gc
xxsimgN = temp.file()
xxsimgE = temp.file()
xxsimgE0 = temp.file()
param = {}
param["i"] = self.iFile
if self.idFN:
param["tid"] = self.idFN
if self.itemFN:
param["item"] = self.itemFN
if self.sp1:
param["s"] = self.sp1
if self.sp2:
param["S"] = self.sp2
#####################
# 異なる向きのconfidenceを列挙するためにsim=C th=0として双方向列挙しておく
# 出力データは倍になるが、mfriendsで-directedとすることで元が取れている
param["sim"] = "C"
param["th"] = "0"
param["node_support"] = True
if self.numtp:
param["num"] = True
param["no"] = xxsimgN
param["eo"] = xxsimgE0
nt.mtra2gc(**param).run()
f = nm.readcsv(xxsimgE0)
for i in range(self.filterSize):
f <<= nm.mselnum(f=self.filter[i],
c="[%s,%s]" % (self.lb[i], self.ub[i]))
f <<= nm.writecsv(xxsimgE)
f.run()
### mfrirends
xxfriends = temp.file()
xxfriendE = temp.file()
xxw = temp.file()
xxf = temp.file()
xxff = temp.file()
xxor = temp.file()
if not os.path.isdir(xxfriends):
os.makedirs(xxfriends)
col = [["FF000080", "FF888880"], ["0000FF80", "8888FF80"],
["00FF0080", "88FF8880"]]
for i in range(len(self.sim)):
paramf = {}
paramf["ei"] = xxsimgE
paramf["ni"] = xxsimgN
paramf["ef"] = "node1,node2"
paramf["nf"] = "node"
paramf["eo"] = xxfriendE
paramf["no"] = xxfriends + "/n_" + str(i)
paramf["sim"] = self.sim[i]
paramf["dir"] = self.dir[i]
paramf["rank"] = self.rank[i]
paramf["directed"] = True
nt.mfriends(**paramf).run()
frec2 = nm.mfsort(f="node1,node2", i=xxfriendE)
frec2 <<= nm.msummary(k="node1,node2",
f=self.sim[i],
c="count,mean")
frec2 <<= nm.mselstr(f="count", v=2)
# node1%0,node2%1,fld,count,mean
# a,b,support,2,0.1818181818
# a,d,support,2,0.1818181818
f = nm.mjoin(k="node1,node2",
K="node1,node2",
m=frec2,
f="mean:s1",
n=True,
i=xxfriendE)
f <<= nm.mjoin(k="node2,node1",
K="node1,node2",
m=frec2,
f="mean:s2",
n=True)
# 1) xxrecs2でsimをjoinできない(s1,s2共にnull)ということは、それは片方向枝なので"F"をつける
# 2) 双方向枝a->b,b->aのうちa->bのみ(s1がnullでない)に"W"の印をつける。
# 3) それ以外の枝は"D"として削除
f <<= nm.mcal(
c='if(isnull($s{s1}),if(isnull($s{s2}),\"F\",\"D\"),\"W\")',
a="dir")
f <<= nm.mselstr(f="dir", v="D", r=True)
f <<= nm.mcal(c='if($s{dir}==\"W\",$s{s1},$s{%s})' % (self.sim[i]),
a="sim")
f <<= nm.mchgstr(f="dir:color",
c='W:%s,F:%s' % (col[i][0], col[i][1]),
A=True)
f <<= nm.msetstr(v=[self.sim[i], str(i)], a="simType,simPriority")
f <<= nm.mcut(f="simType,simPriority,node1,node2,sim,dir,color",
o=xxfriends + "/e_" + str(i))
f.run()
# node1%1,node2%0,simType,sim,dir,color
# b,a,jaccard,0.3333333333,F,8888FF
# j,c,jaccard,0.3333333333,F,8888FF
# b,d,jaccard,0.3333333333,F,8888FF
# a,e,jaccard,0.5,W,0000FF
# d,e,jaccard,0.5,W,0000FF
# rule fileの出力
if self.orFile:
mmm = nm.mcat(i=xxfriends + "/e_*").muniq(k="node1,node2")
nm.mcommon(k="node1,node2", i=xxsimgE, m=mmm, o=self.orFile).run()
# マルチ枝の単一化(W優先,パラメータ位置優先)
if self.prune:
"""
# 双方向と片方向に分割
nm.mcat(i=xxfriends+"/e_*").mselstr(f="dir",v="W",o=xxw,u=xxf).run()
# 片方向のみの枝を選択
f = nm.mcommon(k="node1,node2",K="node1,node2",r=True,m=xxw,i=xxf)
f <<= nm.mcommon(k="node1,node2",K="node2,node1",r=True,m=xxw,o=xxff)
f.run()
f = nm.mcat(i=xxw+","+xxff).mbest(k="node1,node2",s="dir%r,simPriority%n",o=self.oeFile).run()
"""
#これだめ
fo = nm.mcat(i=xxfriends + "/e_*").mselstr(f="dir", v="W")
fu = fo.direction("u") # これは再考
fu <<= nm.mcommon(k="node1,node2", K="node1,node2", r=True, m=fo)
fu <<= nm.mcommon(k="node1,node2", K="node2,node1", r=True, m=fo)
#f = nm.m2cat()
f = nm.mbest(i=[fo, fu],
k="node1,node2",
s="dir%r,simPriority%n",
o=self.oeFile)
f.run()
else:
nm.mcat(i=xxfriends + "/e_*", o=self.oeFile).run()
nm.mcat(i=xxfriends + "/n_0", o=self.onFile).run()