def trainTest(self,testSplit,**kw):
idLab = self.idLab
spDir = self.getSplitDirName(testSplit)
makedir(spDir)
idLabRec = idLab.getRecords()
spIdLab = idLabRec.copy()
spIdLab["split"] = 1
spIdLab["split"][idLabRec["split"]==testSplit] = 2
spIdLabFile = pjoin(spDir,"idlab.pkl")
saveIdLabelRecords(records=spIdLab,fileName=spIdLabFile)
spClOpt = copy(self.clOpt)
spClOpt.labels = [spIdLabFile]
spClOpt.predFile = pjoin(spDir,"pred.pkl")
spClOpt.perfFile = pjoin(spDir,"perf.pkl")
spClOpt.mode = "trainScatter"
spClOpt.runMode = self.opt.runMode
spClOpt.cwd = spDir
spOptFile = self.getSplitOptFileName(testSplit)
spApp = ClassifierApp(opt=spClOpt)
spClOpt = spApp.getOpt() #get the missing options filled with defaults
#that just saves a copy for us - App will create its own when it
#submits a batch job (unique and slightly modified)
dumpObj(spClOpt,spOptFile)
kw = kw.copy()
jobs = spApp.run(**kw)
spClOpt.mode = "test"
spApp = ClassifierApp(opt=spClOpt)
kw["depend"] = jobs
jobs = spApp.run(**kw)
return jobs
def run(self, **kw):
from MGT.ClassifierApp import ClassifierApp
opt = self.opt
opt.cwd = self.path # in case we moved it after __init__()
opt.predFile = pjoin(opt.cwd, "pred.pkl")
jobs = []
opt.mode = "predict"
app = ClassifierApp(opt=opt)
opt = app.getOpt() # get the missing options filled with defaults
jobs = app.run(**kw)
self.save()
return jobs
def run(self, **kw):
from MGT.ClassifierApp import ClassifierApp
opt = self.opt
opt.cwd = self.path # in case we moved it after __init__()
jobs = []
opt.mode = "trainScatter"
app = ClassifierApp(opt=opt)
opt = app.getOpt() # get the missing options filled with defaults
# we need to clean up the models directory otherwise we run a danger
# of picking up old models that are not trained in this session
rmrf(self.getFilePath(opt.modelRoot))
return app.run(**kw)
def makeClOpt(testDataDir):
o,a = ClassifierApp.defaultOptions()
o.C = 0.03
o.kernel = "lin" #"rbf"
origFeat = [pjoin(testDataDir,"usps-train.pkl"),pjoin(testDataDir,"usps-test.pkl")]
o.inFeat = origFeat
#o.inFeat = [pjoin(testDataDir,"usps.libsvm.pca")]
o.labels = [ lf+".idlab" for lf in origFeat ]
return o
def makeClOpt(testDataDir):
o,a = ClassifierApp.defaultOptions()
o.method = "svm" #"svm" #"smlr"
o.C = 0.001
o.kernel = "rbf" #"rbf" "lin"
o.inFeatFormat = "pkl" #"txt" #"pkl"
if o.inFeatFormat == "txt":
ext = ".libsvm"
elif o.inFeatFormat == "pkl":
ext = ".pkl"
else:
raise ValueError(o.inFeatFormat)
origFeat = [pjoin(testDataDir,"usps-train"+ext),pjoin(testDataDir,"usps-test"+ext)]
o.inFeat = origFeat
#o.inFeat = [pjoin(testDataDir,"usps.libsvm.pca")]
o.labels = [ lf+".idlab" for lf in origFeat ]
o.balanceTrainCounts=-1
return o
def run(self, **kw):
from MGT.ClassifierApp import ClassifierApp
opt = self.opt
opt.cwd = self.path # in case we moved it after __init__()
opt.predFile = pjoin(opt.cwd, "pred.pkl")
opt.perfFile = pjoin(opt.cwd, "perf.pkl")
jobs = []
opt.mode = "trainScatter"
app = ClassifierApp(opt=opt)
opt = app.getOpt() # get the missing options filled with defaults
# we need to clean up the models directory otherwise we run a danger
# of picking up old models that are not trained in this session
rmrf(self.getFilePath(opt.modelRoot))
jobs = app.run(**kw)
opt.mode = "test"
app = ClassifierApp(opt=opt)
opt = app.getOpt() # get the missing options filled with defaults
self.save()
kw = kw.copy()
kw["depend"] = jobs
jobs = app.run(**kw)
return jobs
def makeClOpt():
o,a = ClassifierApp.defaultOptions()
o.C = 0.03
o.inFeat = [os.path.abspath("usps-train.libsvm"),os.path.abspath("usps-test.libsvm")]
o.labels = [ lf+".idlab" for lf in o.inFeat ]
return o
### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ## # # See COPYING file distributed along with the MGTAXA package for the # copyright and license terms. # ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ## """Shell scripting interface to supervised classifiers.""" from MGT.ClassifierApp import ClassifierApp app = ClassifierApp() app.run()
else:
raise ValueError(o.inFeatFormat)
origFeat = [pjoin(testDataDir,"usps-train"+ext),pjoin(testDataDir,"usps-test"+ext)]
o.inFeat = origFeat
#o.inFeat = [pjoin(testDataDir,"usps.libsvm.pca")]
o.labels = [ lf+".idlab" for lf in origFeat ]
o.balanceTrainCounts=-1
return o
testDataDir = pjoin(options.testDataDir,"usps")
workDir = pjoin(os.getcwd(),"test_ClassifierApp.tmpdir")
rmdir(workDir)
makedir(workDir)
os.chdir(workDir)
clOpt = makeClOpt(testDataDir)
clOpt.numTrainJobs = 1 #6
clOpt.MEM = 500
clOpt.LENGTH = "fast"
clOpt.mode = "trainScatter"
clOpt.runMode = "inproc" #"inproc" #"batchDep"
print clOpt
app = ClassifierApp(opt=clOpt)
jobs = app.run(dryRun=False)
clOpt.mode = "test"
clOpt.perfFile = "perf.pkl"
app = ClassifierApp(opt=clOpt)
jobs = app.run(dryRun=False,depend=jobs)
