def main():
util.enable_reversecomplement()
args = loadargs()
models = loadmodels(args)
trdata = None
tedata = None
tfids = load_tfids(args)
for tfid in tfids:
if "calib" in args.steps:
print "-------------- calib:", tfid, "--------------"
trdata = load_traindata(tfid, args)
util.calibrate(models, trdata, args.calibdir, nfold=args.nfold, ncalib=args.ncalib)
if "train" in args.steps:
print "-------------- train:", tfid, "--------------"
trdata = load_traindata(tfid, args)
util.train(models, trdata, args.calibdir, args.finaldir, nfold=1, ntrial=args.ntrial)
if "test" in args.steps:
tedata = load_testdata(tedata, tfids, args)
util.save_metrics(tedata, "test", args.finaldir)
if "report" in args.steps:
tedata = load_testdata(tedata, tfids, args)
util.save_featuremaps(tedata, args.finaldir, args.reportdir, maxrows=100000)
util.save_report(args.finaldir, args.reportdir, tfids)
def main():
util.enable_reversecomplement()
args = loadargs()
models = loadmodels(args)
tfgroups = load_tfgroups(args)
util.globals.flags.push("normalize_targets", True)
for tfgroup in tfgroups:
trdata = None
if len(tfgroup["ids"]) == 0:
print "No TFs to train on microarray %s"%tfgroup["train_fold"]
continue
if "calib" in args.steps:
trdata = load_pbmdata(trdata, tfgroup["ids"], tfgroup["train_fold"], args, remove_probe_bias=True)
util.calibrate(models, trdata, args.calibdir, nfold=args.nfold, ncalib=args.ncalib, allfolds=True)
if "train" in args.steps:
trdata = load_pbmdata(trdata, tfgroup["ids"], tfgroup["train_fold"], args, remove_probe_bias=True)
util.train(models, trdata, args.calibdir, args.finaldir, nfold=1, ntrial=args.ntrial, metric_key="pearson.r")
for tfgroup in tfgroups:
tedata = None
newids = []
for id in tfgroup["ids"]:
if os.path.exists(args.outdir+"/final/"+id+"/model.pkl"):
newids.append(id)
else:
print "WARNING: did not find model for %s, skipping" % id
tfgroup["ids"] = newids
if len(tfgroup["ids"]) == 0:
print "No TFs to test on microarray %s"%tfgroup["train_fold"]
continue
if "test" in args.steps:
tedata = load_pbmdata(tedata, tfgroup["ids"], tfgroup["test_fold"], args, remove_probe_bias=False)
save_test_performance(tedata, tfgroup["ids"], tfgroup["test_fold"], args)
if "report" in args.steps:
tedata = load_pbmdata(tedata, tfgroup["ids"], tfgroup["test_fold"], args, remove_probe_bias=False)
util.save_featuremaps(tedata, args.finaldir, args.reportdir)
if "report" in args.steps:
all_tfids = sum([tfgroup["ids"] for tfgroup in tfgroups],[])
save_report(args.finaldir, args.reportdir, all_tfids, index_metric="pearson")
save_pbm_performance_table(args, all_tfids)
if "chip" in args.steps:
all_tfids = sum([tfgroup["ids"] for tfgroup in tfgroups],[])
save_chip_performance_table(args, all_tfids)
def main():
util.enable_reversecomplement()
args = loadargs()
models = loadmodels(args)
trdata = None
tedata = None
tfids = load_tfids(args)
for tfid in tfids:
if "calib" in args.steps:
print "-------------- calib:", tfid, "--------------"
set_motif_lengths(args, models, tfid)
trdata = load_traindata(tfid, args)
util.calibrate(models,
trdata,
args.calibdir,
nfold=args.nfold,
ncalib=args.ncalib,
allfolds=False)
if "train" in args.steps:
print "-------------- train:", tfid, "--------------"
set_motif_lengths(args, models, tfid)
trdata = load_traindata(tfid, args)
util.train(models,
trdata,
args.calibdir,
args.finaldir,
nfold=1,
ntrial=args.ntrial)
if "test" in args.steps:
tedata = load_testdata(tedata, tfids, args)
util.save_metrics(tedata, "test", args.finaldir)
if "report" in args.steps:
tedata = load_testdata(tedata, tfids, args)
util.save_featuremaps(tedata, args.finaldir, args.reportdir)
util.save_report(args.finaldir, args.reportdir, tfids)
def main():
args = loadargs()
models = loadmodels(args)
trdata = None
util.globals.flags.push(
"clamp_targets", True
) # Clamp the top .995 percentile of target values, to avoid extremely large targets (e.g. 00038/MBNL suffers from this)
util.globals.flags.push("normalize_targets",
True) # Make the targets have unit variance
if "calib" in args.steps:
trdata = load_traindata(trdata, args)
util.calibrate(models,
trdata,
args.calibdir,
nfold=args.nfold,
ncalib=args.ncalib,
allfolds=True)
if "train" in args.steps:
trdata = load_traindata(trdata, args)
util.train(models,
trdata,
args.calibdir,
args.finaldir,
nfold=1,
ntrial=args.ntrial)
if "test" in args.steps:
save_test_predictions(args)
if "report" in args.steps:
save_pfms(args)
ids, _ = get_chunktargets(args)
ids = sorted(ids)
util.save_report(args.finaldir,
args.reportdir,
ids,
index_metric="pearson",
rna=True)
def main():
util.enable_reversecomplement()
args = loadargs()
models = loadmodels(args)
tfgroups = load_tfgroups(args)
util.globals.flags.push("normalize_targets", True)
for tfgroup in tfgroups:
trdata = None
if len(tfgroup["ids"]) == 0:
print "No TFs to train on microarray %s" % tfgroup["train_fold"]
continue
if "calib" in args.steps:
trdata = load_pbmdata(trdata,
tfgroup["ids"],
tfgroup["train_fold"],
args,
remove_probe_bias=True)
util.calibrate(models,
trdata,
args.calibdir,
nfold=args.nfold,
ncalib=args.ncalib,
allfolds=True)
if "train" in args.steps:
trdata = load_pbmdata(trdata,
tfgroup["ids"],
tfgroup["train_fold"],
args,
remove_probe_bias=True)
util.train(models,
trdata,
args.calibdir,
args.finaldir,
nfold=1,
ntrial=args.ntrial,
metric_key="pearson.r")
for tfgroup in tfgroups:
tedata = None
newids = []
for id in tfgroup["ids"]:
if os.path.exists(args.outdir + "/final/" + id + "/model.pkl"):
newids.append(id)
else:
print "WARNING: did not find model for %s, skipping" % id
tfgroup["ids"] = newids
if len(tfgroup["ids"]) == 0:
print "No TFs to test on microarray %s" % tfgroup["train_fold"]
continue
if "test" in args.steps:
tedata = load_pbmdata(tedata,
tfgroup["ids"],
tfgroup["test_fold"],
args,
remove_probe_bias=False)
save_test_performance(tedata, tfgroup["ids"], tfgroup["test_fold"],
args)
if "report" in args.steps:
tedata = load_pbmdata(tedata,
tfgroup["ids"],
tfgroup["test_fold"],
args,
remove_probe_bias=False)
util.save_featuremaps(tedata, args.finaldir, args.reportdir)
if "report" in args.steps:
all_tfids = sum([tfgroup["ids"] for tfgroup in tfgroups], [])
save_report(args.finaldir,
args.reportdir,
all_tfids,
index_metric="pearson")
save_pbm_performance_table(args, all_tfids)
if "chip" in args.steps:
all_tfids = sum([tfgroup["ids"] for tfgroup in tfgroups], [])
save_chip_performance_table(args, all_tfids)