def train(self):
self.trainData.flush()
cmd = "%s -fvals %s %s" % (megam, self.mode, self.trainData.name)
err = StringIO()
out = StringIO()
bettersystem(cmd, stdout=out, stderr=err)
self.trainData.close()
self.trainData = None
#print err.getvalue()
#print out.getvalue()
self.readModel(out.getvalue())
def train(self):
self.trainData.flush()
cmd = "%s -fvals %s %s" % (megam, self.mode, self.trainData.name)
err = StringIO()
out = StringIO()
bettersystem(cmd, stdout=out, stderr=err)
self.trainData.close()
self.trainData = None
#print err.getvalue()
#print out.getvalue()
self.readModel(out.getvalue())
def runtask(models, datadir):
mfile = datadir/"matrix"
bin = myBin(datadir)
solver = bin/"chainedSolvers"
mfile = datadir/"matrix"
outfile = datadir/("output-%s" % "-".join(models[0].split()))
output = StringIO()
cmd = "%s %s %s" % (solver, " ".join(models), mfile)
print "Running", cmd
bettersystem(cmd, stdout=output)
print "Done, postprocessing output..."
outfh = file(outfile, 'w')
for line in output.getvalue().split("\n"):
if not line.strip().isdigit():
print line
else:
print >>outfh, line
outfh.close()
return ResultsFile(outfile)
def run(self, hogwash_job):
on64 = (get_cpu_bitness() == 64)
if on64:
par = self.solverCmd.parent
self.solverCmd = par/"../bin64"/self.solverCmd.basename()
outfile = make_job_output_filename(hogwash_job, "output")
cmd = "%s %s %s > %s" % (self.solverCmd, " ".join(self.solvers),
self.mfile, outfile)
print cmd
status = bettersystem(cmd)
if status != 0:
raise BadExitCode(status)
return ResultsFile(outfile)
devlog = file(modeldir / "devkeys", 'w')
devfeats = file(modeldir / "devfeats", 'w')
conditionalFeatures.timeSpanFeats(dev,
logfile=devlog,
blocksize=blocksize,
output=devfeats)
devlog.close()
devfeats.close()
trainErr = StringIO()
bettersystem(megam + " -fvals binary " + modeldir / "feats" + " > " +
modeldir / "model",
stderr=trainErr)
testErr = StringIO()
bettersystem(megam + " -predict " + modeldir / "model"
" -fvals binary " + modeldir / "devfeats" + "> " +
modeldir / "predictions",
stderr=testErr)
trainErrStr = lastLine(trainErr)
testErrStr = lastLine(testErr)
print "Blocksize", blocksize, \
"train err", trainErrStr, "test err", testErrStr
print "Majority baseline error", \
false1 += 1
elif truth == 1:
missed1 += 1
lines += 1
prec = 0
if true1 + false1 > 0:
prec = true1 / (true1 + false1)
rec = true1 / (true1 + missed1)
f = 0
if prec + rec > 0:
f = (2*prec*rec)/(prec+rec)
err = (lines-right)/lines
acc = 1 - err
print "Acc", acc, "P", prec, "R", rec, "F", f
if __name__ == "__main__":
blockdir = path(argv[1])
model = blockdir/"model"
feats = blockdir/"devfeats"
predictions = StringIO()
bettersystem(megam + " -predict " + model +
" -fvals binary " + feats, stdout=predictions,
stderr=StringIO())
score(predictions, feats)
print "True clustering has", \
len(set([pt.label for pt in test])), "clusters"
print "Objective value of truth: %.3g" % objective(test, clMat)
print
for algorithm in ["first", "best", "vote", "pivot"]:
evals = []
print "================", algorithm, "================="
print
for run in range(10):
cmd = "%s -a %s %s" % (tester, algorithm, classFileName)
output = StringIO()
status = bettersystem(cmd, stdout=output, stderr=StringIO())
assert(status == 0)
prop = readDataFile(output.getvalue().split("\n"))
score = Eval(test, prop, clMat, filename=("run%d" % run))
evals.append(score)
print "Run", run
print score
print algorithm
print "Best Objective:"
print min(evals, key=lambda x: x.stats["objective"])
print "Average:"
print reduce(lambda x,y: x+y, evals).normalized(len(evals))
for power in range(1,4):
for layer in range(9):
dsize = (layer + 1) * 10**power
print "Data: size", dsize
if dsize >= 5000:
break
for run in range(3):
print "Run:", run
ddir = workdir/("data%d-%d" % (dsize, run))
ddir.mkdir()
output = StringIO()
cmd = "python script/genGaussians.py -n %d -t 10 -f 3 -a 1"
bettersystem(cmd % dsize, stdout=output)
# bettersystem("python script/genFeatureless.py -n %d" %
# dsize, stdout=output)
dfileName = ddir/"data"
dfile = file(dfileName, 'w')
dfile.write(output.getvalue()),
dfile.close()
output = StringIO()
bettersystem("python script/classify.py -c max-ent %s" %
dfileName, stdout=output)
# bettersystem("python script/classify.py %s" %
# dfileName, stdout=output)
