def main(args):
# assertions
assert args.criterion in ['gini', 'entropy']
# create output dir
out_dir = os.path.join(args.out_dir,
args.dataset,
args.criterion,
'rs_{}'.format(args.rs),
'topd_{}'.format(args.topd),
'k_{}'.format(args.k),
'sub_{}'.format(args.subsample_size))
# create output directory and clear previous contents
os.makedirs(out_dir, exist_ok=True)
print_util.clear_dir(out_dir)
# skip experiment if results already exist
if args.append_results and os.path.exists(os.path.join(out_dir, 'results.npy')):
print('results exist: {}'.format(out_dir))
return
# create logger
log_fp = os.path.join(out_dir, 'log.txt')
logger = print_util.get_logger(log_fp)
logger.info(args)
logger.info(datetime.now())
# run experiment
experiment(args, logger, out_dir, seed=args.rs)
# remove logger
print_util.remove_logger(logger)
def main(args):
# make logger
dataset = args.dataset
if args.train_frac < 1.0 and args.train_frac > 0.0:
dataset += '_{}'.format(str(args.train_frac).replace('.', 'p'))
out_dir = os.path.join(args.out_dir, dataset, args.tree_type,
'rs{}'.format(args.rs))
if args.trex:
out_dir = os.path.join(out_dir, args.kernel_model, args.tree_kernel)
elif args.teknn:
out_dir = os.path.join(out_dir, 'teknn', args.tree_kernel)
elif args.maple:
out_dir = os.path.join(out_dir, 'maple')
elif args.inf_k is not None:
out_dir = os.path.join(out_dir, 'leaf_influence')
elif args.mmd:
out_dir = os.path.join(out_dir, 'mmd')
elif args.proto:
out_dir = os.path.join(out_dir, 'proto')
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info(datetime.now())
seed = args.rs
logger.info('\nSeed: {}'.format(seed))
experiment(args, logger, out_dir, seed=seed)
print_util.remove_logger(logger)
def main(args):
# make logger
dataset = args.dataset
for rs in args.rs:
out_dir = os.path.join(args.out_dir, dataset, args.tree_type,
'rs{}'.format(rs))
if args.trex:
out_dir = os.path.join(out_dir, args.kernel_model, args.tree_kernel)
elif args.teknn:
out_dir = os.path.join(out_dir, 'teknn', args.tree_kernel)
elif args.maple:
out_dir = os.path.join(out_dir, 'maple')
elif args.inf_k is not None:
out_dir = os.path.join(out_dir, 'leaf_influence')
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('\nSeed: {}'.format(rs))
experiment(args, logger, out_dir, seed=rs)
print_util.remove_logger(logger)
def main(args):
# create output dir
out_dir = os.path.join(args.out_dir,
args.dataset,
args.criterion,
args.method,
'rs_{}'.format(args.rs))
log_fp = os.path.join(out_dir, 'log.txt')
os.makedirs(out_dir, exist_ok=True)
# skip experiment if results already exist
if args.append_results and os.path.exists(os.path.join(out_dir, 'results.npy')):
return
# create logger
logger = print_util.get_logger(log_fp)
logger.info(args)
logger.info(datetime.now())
# run experiment
experiment(args, logger, out_dir)
# remove logger
print_util.remove_logger(logger)
def main(args):
out_dir = os.path.join(args.out_dir)
# create logger
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info(datetime.now())
create_csv(args, out_dir, logger)
def main(args):
# make logger
dataset = args.dataset
out_dir = os.path.join(args.out_dir, dataset, args.model)
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, '{}.txt'.format(args.dataset)))
logger.info(args)
experiment(args, logger, out_dir, seed=args.rs)
def main(args):
# make logger
dataset = args.dataset
out_dir = os.path.join(args.out_dir, dataset, args.tree_type, args.tree_kernel)
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
experiment(args, logger, out_dir, seed=args.rs)
print_util.remove_logger(logger)
def main(args):
# create output directory
out_dir = os.path.join(args.out_dir, args.dataset)
os.makedirs(out_dir, exist_ok=True)
# create logger
logger_fp = os.path.join(out_dir, 'log.txt')
logger = print_util.get_logger(logger_fp)
logger.info('{}'.format(args))
logger.info('\ntimestamp: {}'.format(datetime.now()))
# get dataset
X_train, X_test, y_train, y_test = data_util.get_data(
args.dataset, args.data_dir)
logger.info('X_train.shape: {}'.format(X_train.shape))
# collect top threshold scores
top_scores = []
# get best threshold(s) for each feature
for i in range(X_train.shape[1]):
vals = np.unique(X_train[:, i])
C = get_thresholds(X_train[:, i], y_train)
S = compute_scores(C)
logger.info(
'\n[FEATURE {}] no. unique: {:,}, no. valid thresholds: {:,}'.
format(i, len(vals), len(C)))
# sort thresholds based on score
S = sorted(S, key=lambda x: x[1])
# display split score for each threshold
for T, s in S[:args.k]:
logger.info(' threshold value: {:.5f}, score: {:.5f}'.format(
T.v, s))
top_scores.append(s)
# plot distribution of top threshold scores
ax = sns.distplot(top_scores, rug=True, hist=False)
ax.set_title('{}: Scores for Top {} Threshold(s) / Feature'.format(
args.dataset.title(), args.k))
ax.set_xlabel('Gini index')
ax.set_ylabel('Density')
plt.savefig(os.path.join(out_dir, 'k_{}.pdf'.format(args.k)),
bbox_inches='tight')
def main(args):
# create output dir
out_dir = os.path.join(args.out_dir, args.dataset, args.criterion)
# add tuning to filepath
if args.no_tune:
out_dir = os.path.join(out_dir, 'no_tune', 'rs_{}'.format(args.rs))
else:
out_dir = os.path.join(out_dir, 'tuned', 'rs_{}'.format(args.rs))
# create filename
if args.model == 'sklearn':
out_dir = os.path.join(out_dir, args.model)
if args.bootstrap:
out_dir = os.path.join(out_dir, 'bootstrap')
elif args.model == 'dare':
assert args.topd == 0
out_dir = os.path.join(out_dir, args.model)
elif args.model in ['extra_trees', 'extra_trees_k1', 'borat']:
out_dir = os.path.join(out_dir, args.model)
else:
raise ValueError('model {} unknown!'.format(args.model))
# create output directory and clear any previous contents
os.makedirs(out_dir, exist_ok=True)
print_util.clear_dir(out_dir)
# create logger
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info(datetime.now())
# write everything printed to stdout to this log file
logfile, stdout, stderr = print_util.stdout_stderr_to_log(
os.path.join(out_dir, 'log+.txt'))
# run experiment
performance(args, out_dir, logger)
# restore original stdout and stderr settings
print_util.reset_stdout_stderr(logfile, stdout, stderr)
def main(args):
# make logger
dataset = args.dataset
for i in range(args.repeats):
seed = args.rs + i
rs_dir = os.path.join(args.out_dir, dataset, args.tree_type,
args.tree_kernel, 'rs{}'.format(seed))
os.makedirs(rs_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(rs_dir, 'log.txt'))
logger.info(args)
logger.info('Seed {}'.format(seed))
experiment(args, logger, rs_dir, seed=seed)
print_util.remove_logger(logger)
def main(args):
# create output dir
out_dir = os.path.join(args.out_dir, args.dataset, args.criterion,
'rs_{}'.format(args.rs))
# create output directory and clear previous contents
os.makedirs(out_dir, exist_ok=True)
print_util.clear_dir(out_dir)
# create logger
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info('timestamp: {}'.format(datetime.now()))
# run experiment
performance(args, out_dir, logger)
def main(args):
# create output dir
out_dir = os.path.join(args.out_dir, args.dataset, args.model_type)
if args.no_tune:
out_dir = os.path.join(out_dir, 'trees_{}'.format(args.n_estimators),
'depth_{}'.format(args.max_depth))
os.makedirs(out_dir, exist_ok=True)
# create logger
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
logger.info(datetime.now())
# run experiment
performance(args, logger)
def main(args):
# make logger
dataset = args.dataset
out_dir = os.path.join(args.out_dir, dataset, args.tree_type,
args.tree_kernel)
if args.trex:
out_dir = os.path.join(out_dir, args.kernel_model)
elif args.teknn:
out_dir = os.path.join(out_dir, 'teknn')
os.makedirs(out_dir, exist_ok=True)
logger = print_util.get_logger(os.path.join(out_dir, 'log.txt'))
logger.info(args)
seed = args.rs
logger.info('\nSeed: {}'.format(seed))
experiment(args, logger, out_dir, seed=seed)
print_util.remove_logger(logger)
