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__()
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 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__()
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