def parser(def_model='l2svm', def_loss='f1score', def_verbose=True):
usage = "usage: %prog [options] arg"
parser = OptionParser(usage)
parser.add_option("-m", "--model", dest="model", default='l2 square', type='string',
#choices=known_models,
help="model selection from {%s}"%', '.join(s for s in known_models))
parser.add_option("-i", "--indir", dest="indir",
help="input directory", metavar="INDIR")
parser.add_option("-o", "--outdir", dest="outdir",
help="output directory", metavar="OUTDIR")
parser.add_option("-l", "--lscore", dest="scorer", default='hamming', type='choice',
choices=['hamming', 'rankloss'], help="score type, hamming or rankloss (def=rankloss)")
#parser.add_option("-s", "--shuffle", dest="shuffle", default=False,
# action="store_true", help="shuffle labels (for random baseline)")
parser.add_option("-t", "--toy", default="0", type="choice",
choices=["0", "1"], dest="istoy",
help='generate toy data instead of real data (-t 1) or no (-t 0)' )
parser.add_option("-q", "--quiet", default=False,
action="store_true", dest="quiet")
(options, args) = parser.parse_args()
# extract options
model_all = options.model # model type
#shuffle = options.shuffle # shuffle yes/no
verbose = not options.quiet
scorer = options.scorer
istoy = int(options.istoy) # yes/no running smaller test data
if istoy:
cv_dir = os.path.expanduser(os.path.join("~", "workdir", "multilabel", "toy", "results", "doublecv"))
else:
cv_dir = os.path.expanduser(os.path.join("~", "workdir", "multilabel", "results", "doublecv"))
resultdir = os.path.join(cv_dir, "processed")
#if shuffle:
# cv_dir = os.path.join(cv_dir, "shuffle")
if options.indir is not None:
cv_dir = options.indir
if options.outdir is not None:
resultdir = options.outdir
# make result directory
try:
os.makedirs(resultdir)
except:
pass
# extract model name
model = clean_name(model_all)
saveinfo = {'resultdir':resultdir}
runinfo = {'model':model, "cv_dir":cv_dir, "scorer":scorer}#, "shuffle":shuffle}
return saveinfo, runinfo, verbose
def main():
usage = "usage: %prog [options] arg"
parser = OptionParser(usage)
parser.add_option("-m", "--model", dest="model", default='l2 svm', type='string',
#choices=known_models,
help="model selection from {%s}"%', '.join(s for s in known_models))
parser.add_option("-i", "--indir", dest="indir",
help="input directory", metavar="INDIR")
parser.add_option("-o", "--outdir", dest="outdir",
help="output directory", metavar="OUTDIR")
parser.add_option("-c", "--cv", dest="CV", type="int", default=0,
help="which CV set (typically 0...4)")
parser.add_option("-l", "--lscore", dest="scorer", default='hamming', type='choice',
choices=['hamming', 'rankloss'], help="score type, hamming or rankloss (def=rankloss)")
parser.add_option("-p", dest="param", type=float, default=10,
help="number of model hyperparameter to test (def=10)")
parser.add_option("-n", "--n_jobs", dest="n_jobs", type=int, default=6,
help="number of processors used in parallel fit (def=6)")
parser.add_option("-s", "--shuffle", dest="shuffle", default=False,
action="store_true", help="shuffle labels (for random baseline)")
parser.add_option("-t", "--toy", default="1", type="choice",
choices=["0", "1"], dest="istoy",
help='generate toy data instead of real data (-t 1) or no (-t 0)' )
parser.add_option("-q", "--quiet", default=False,
action="store_true", dest="quiet")
(options, args) = parser.parse_args()
# extract options
model_all = options.model # model type
cv = options.CV # which cv set
p = options.param # number of params
shuffle = options.shuffle # shuffle yes/no
verbose = 0 if options.quiet else 1
scorer = options.scorer
istoy = int(options.istoy) # yes/no running smaller test data
n_jobs = options.n_jobs # number f parallel fit jobs
datadir = os.path.expanduser(os.path.join("~", "workdir", "data", "openfMRI", "preprocessed"))
if istoy:
resultdir = os.path.expanduser(os.path.join("~", "workdir", "multilabel", "toy", "results", "wardcv"))
else:
resultdir = os.path.expanduser(os.path.join("~", "workdir", "multilabel", "results", "wardcv"))
if shuffle:
resultdir = os.path.join(resultdir, "shuffle")
if options.indir is not None:
datadir = options.indir
if options.outdir is not None:
resultdir = options.outdir
# make result directory
try:
os.makedirs(resultdir)
except:
pass
# extract model name
model = clean_name(model_all)
# run it!
###################################################
# extract_data
if verbose:
print "Extracting data..."
T = time.time()
Xt, Xe, Z, train, test, subtrain, subtest = extract_data(datadir, cv, verbose, istoy)
if verbose:
dims = Xt.shape+(Z.shape[1],)
print "data dims: N=%d, D=%d, K=%d"%dims
if verbose:
print "extract time = %s"%(time.time()-T,)
if shuffle: print "shuffling output labels..."
if verbose:
print "Training model...\nmodel=%s, cv=%d, scorer=%s"%(model, cv, scorer)
T = time.time()
# run model (or multimodel).
saveinfo = {'resultdir':resultdir, 'cv':cv}
runinfo = {'model':model, "scorer":scorer, "p":p, "shuffle":shuffle, "n_jobs":n_jobs}
data = (Xt, Xe, Z, train, test, subtrain, subtest)
train_multilabel(data, runinfo, saveinfo, verbose)
if verbose: print "Combined Train time = %s"%(time.time()-T,)