def main(opt):
""" Main routine that exits with status code 0
"""
### pre process
# load model file
clr = pickle.load(opt.model)
clr_info = pickle.load(opt.model)
# read data
D = np.loadtxt(opt.infile)
# split data and process missing values
y = np.array(D[:, -1])
if opt.ns:
X = fill_missing_with_mean(D[:, :-(1 + N_NS)])
else:
X = fill_missing_with_mean(D[:, :-1])
S = np.atleast_2d(D[:, -(1 + N_NS):-1])
### main process
# set starting time
start_time = datetime.datetime.now()
start_utime = os.times()[0]
opt.start_time = start_time.isoformat()
logger.info("start time = " + start_time.isoformat())
# prediction and write results
p = clr.predict_proba(X)
# output prediction
n = 0
m = 0
for i in xrange(p.shape[0]):
c = np.argmax(p[i, :])
opt.outfile.write("%d %d " % (y[i], c))
opt.outfile.write(" ".join(S[i, :].astype(str)) + " ")
opt.outfile.write(str(p[i, 0]) + " " + str(p[i, 1]) + "\n")
n += 1
m += 1 if c == y[i] else 0
# set end and elapsed time
end_time = datetime.datetime.now()
end_utime = os.times()[0]
logger.info("end time = " + end_time.isoformat())
opt.end_time = end_time.isoformat()
logger.info("elapsed_time = " + str((end_time - start_time)))
opt.elapsed_time = str((end_time - start_time))
logger.info("elapsed_utime = " + str((end_utime - start_utime)))
opt.elapsed_utime = str((end_utime - start_utime))
### output
# add meta info
opt.nos_samples = n
logger.info('nos_samples = ' + str(opt.nos_samples))
opt.nos_correct_samples = m
logger.info('nos_correct_samples = ' + str(opt.nos_correct_samples))
opt.accuracy = m / float(n)
logger.info('accuracy = ' + str(opt.accuracy))
opt.negative_mean_prob = np.mean(p[:, 0])
logger.info('negative_mean_prob = ' + str(opt.negative_mean_prob))
opt.positive_mean_prob = np.mean(p[:, 1])
logger.info('positive_mean_prob = ' + str(opt.positive_mean_prob))
# output meta information
if opt.info:
for key in clr_info.keys():
opt.outfile.write("#classifier_%s=%s\n" %
(key, str(clr_info[key])))
for key, key_val in vars(opt).iteritems():
opt.outfile.write("#%s=%s\n" % (key, str(key_val)))
### post process
# close file
if opt.infile != sys.stdin:
opt.infile.close()
if opt.outfile != sys.stdout:
opt.outfile.close()
if opt.model != sys.stdout:
opt.model.close()
sys.exit(0)
def main(opt):
""" Main routine that exits with status code 0
"""
### pre process
# read data
D = np.loadtxt(opt.infile)
# split data and process missing values
y = np.array(D[:, -1])
X = fill_missing_with_mean(D[:, :-1])
del D
### main process
# set starting time
start_time = datetime.datetime.now()
start_utime = os.times()[0]
opt.start_time = start_time.isoformat()
logger.info("start time = " + start_time.isoformat())
# init constants
ns = 1
# train
if opt.ntry <= 0:
# train only once with zero coefficients
clr = train(X, y, ns, opt)
opt.final_loss = clr.f_loss_
logger.info('final_loss = ' + str(opt.final_loss))
else:
# train multiple times with random restarts
clr = None
best_loss = np.inf
best_trial = 0
for trial in xrange(opt.ntry):
logger.info("Trial No. " + str(trial + 1))
tmp_clr = train(X, y, ns, opt)
logger.info("loss = " + str(tmp_clr.f_loss_))
if tmp_clr.f_loss_ < best_loss:
clr = tmp_clr
best_loss = clr.f_loss_
best_trial = trial + 1
opt.final_loss = best_loss
logger.info('final_loss = ' + str(opt.final_loss))
opt.best_trial = best_trial
logger.info('best_trial = ' + str(opt.best_trial))
# set end and elapsed time
end_time = datetime.datetime.now()
end_utime = os.times()[0]
logger.info("end time = " + end_time.isoformat())
opt.end_time = end_time.isoformat()
logger.info("elapsed_time = " + str((end_time - start_time)))
opt.elapsed_time = str((end_time - start_time))
logger.info("elapsed_utime = " + str((end_utime - start_utime)))
opt.elapsed_utime = str((end_utime - start_utime))
### output
# add info
opt.nos_samples = X.shape[0]
logger.info('nos_samples = ' + str(opt.nos_samples))
opt.nos_features = X.shape[1]
logger.info('nos_features = ' + str(X.shape[1]))
opt.classifier = clr.__class__.__name__
logger.info('classifier = ' + opt.classifier)
opt.fadm_version = fadm_version
logger.info('fadm_version = ' + opt.fadm_version)
opt.sklearn_version = sklearn_version
logger.info('sklearn_version = ' + opt.sklearn_version)
# opt.training_score = clr.score(X, y)
# logger.info('training_score = ' + str(opt.training_score))
# write file
pickle.dump(clr, opt.outfile)
info = {}
for key, key_val in vars(opt).iteritems():
info[key] = str(key_val)
pickle.dump(info, opt.outfile)
### post process
# close file
if opt.infile is not sys.stdin:
opt.infile.close()
if opt.outfile is not sys.stdout:
opt.outfile.close()
sys.exit(0)
def main(opt):
""" Main routine that exits with status code 0
"""
### pre process
# read data
D = np.loadtxt(opt.infile)
# split data and process missing values
y = np.array(D[:, -1])
if opt.ns:
X = fill_missing_with_mean(D[:, :-(1 + N_NS)])
else:
X = fill_missing_with_mean(D[:, :-1])
### main process
# set starting time
start_time = datetime.datetime.now()
start_utime = os.times()[0]
opt.start_time = start_time.isoformat()
logger.info("start time = " + start_time.isoformat())
# main process
clr = LogisticRegression(C=float(opt.C), penalty='l2', fit_intercept=True)
clr.fit(X, y)
# set end and elapsed time
end_time = datetime.datetime.now()
end_utime = os.times()[0]
logger.info("end time = " + end_time.isoformat())
opt.end_time = end_time.isoformat()
logger.info("elapsed_time = " + str((end_time - start_time)))
opt.elapsed_time = str((end_time - start_time))
logger.info("elapsed_utime = " + str((end_utime - start_utime)))
opt.elapsed_utime = str((end_utime - start_utime))
### output
# add info
opt.nos_samples = X.shape[0]
logger.info('nos_samples = ' + str(opt.nos_samples))
opt.nos_features = X.shape[1]
logger.info('nos_features = ' + str(X.shape[1]))
opt.classifier = clr.__class__.__name__
logger.info('classifier = ' + opt.classifier)
opt.fadm_version = fadm_version
logger.info('fadm_version = ' + opt.fadm_version)
# opt.training_score = clr.score(X, y)
# logger.info('training_score = ' + str(opt.training_score))
# write file
pickle.dump(clr, opt.outfile)
info = {}
for key, key_val in vars(opt).iteritems():
info[key] = str(key_val)
pickle.dump(info, opt.outfile)
### post process
# close file
if opt.infile is not sys.stdin:
opt.infile.close()
if opt.outfile is not sys.stdout:
opt.outfile.close()
sys.exit(0)