def smoke_build_model(build_model):
set_log_file("testlog.log")
prepared_data = prepare_fake_data()
f_train, f_validate, train_idx, valid_idx, train_eval, valid_eval = (
build_model(prepared_data, batch_size=1))
best_validation_loss, best_epoch = (
train_model(f_train, f_validate, train_idx, valid_idx, train_eval, valid_eval,
n_epochs=100))
assert best_validation_loss > 0.5
assert best_epoch > 0
help='the name of the parameter set that we want to use')
parser.add_argument('-f',
dest='file',
type=str,
default=None,
help='the data file to use')
parser.add_argument('-o',
dest='outname',
type=str,
default=gen_log_name(),
help='name for the log file to be generated')
parser.add_argument(
'-t',
dest='task_num',
type=int,
default=0,
help='a way to separate different runs of the same parameter-set')
args = parser.parse_args()
params = config.get_config(args.param_set)
set_log_file(args.outname)
if args.file:
params['dataset_name'] = args.file
elif 'dataset_name' not in params:
params['dataset_name'] = default_dataset
run(0, **params)
print "finished"
if sys.platform.startswith('win'):
from win_utils import winalert
winalert()
dataset = smart_load_data(**kwargs)
train_idx, valid_idx = cv_split_binarized(dataset, percent=0.2, fold_index=task_num)
prepared_data = (dataset, train_idx, valid_idx)
# load classifiers to build ensemble out of
saved_classifiers = filter(lambda(fn): os.path.splitext(fn)[1] == '.model', os.listdir('.'))
# build and train ensemble
model = SelectedModel(prepared_data, saved_classifiers=saved_classifiers, **kwargs)
_, params = model.train_full(**kwargs)
if __name__ == '__main__':
args = docopt(__doc__)
params = {}
log_filename = args['--out'] or gen_log_name()
if args['--quiet']:
log_filename = os.devnull
print("Not printing to log file.")
set_log_file(log_filename)
if args['--file']:
params['dataset_name'] = args['--file']
task_num = int(args['--task_number'])
params['conds'] = COND_TYPES[task_num % len(COND_TYPES)]
run(task_num=task_num, **params)
print("Finished")
with open(fname, 'w') as f:
writer = csv.writer(f, delimiter='\t')
writer.writerow(headers)
for i in idxs:
student = data['subject'][i]
outcome = 'correct' if data['correct'][i] == 2 else 'incorrect'
KCs = 'word' if single_skill else data.orig['skill'][i]
word_feats = skills[data['skill'][i]]
row = [student, outcome, KCs] + list(word_feats)
writer.writerow(row)
def to_fast(data, train_fname=None, valid_fname=None, fold=0, vector_length=150, single_skill=False):
train_fname = train_fname or 'FAST+deepkt_train{}.txt'.format(fold)
valid_fname = valid_fname or 'FAST+deepkt_test{}.txt'.format(fold)
valid_idx, train_idx = cv_split(data, fold_index=fold, no_new_skill=True)
skills = gen_word_matrix(data.get_data('skill'), data['skill'].enum_pairs, vector_length=vector_length)
# create fast header
feat_headers = ['*features_{}'.format(i) for i in xrange(vector_length)]
headers = ['student', 'outcome', 'KCs'] + feat_headers
_to_fast_with_idxs(train_fname, headers, skills, data, train_idx, single_skill=single_skill)
_to_fast_with_idxs(valid_fname, headers, skills, data, valid_idx, single_skill=single_skill)
set_log_file('temp.log')
#for fold in xrange(14):
# to_fast(fold=fold, single_skill=False)
data = prepare_data(dataset_name='data/data5.gz', top_n=14)
to_fast(data, fold=0, single_skill=False)
prepared_data = chinese.prepare_data(top_n=40, **kwargs)
train_idx, valid_idx = cv_split(prepared_data, percent=.1, fold_index=task_num, **kwargs)
model = SelectedModel((prepared_data, train_idx, valid_idx), **kwargs)
model.train_full()
if __name__ == '__main__':
default_dataset = 'raw_data/chinese_dictation.txt'
parser = argparse.ArgumentParser(description="run an experiment on this computer")
parser.add_argument('-p', dest='param_set', type=str, default='default',
choices=config.all_param_set_keys,
help='the name of the parameter set that we want to use')
parser.add_argument('-f', dest='file', type=str, default=None,
help='the data file to use')
parser.add_argument('-o', dest='outname', type=str, default=gen_log_name(),
help='name for the log file to be generated')
parser.add_argument('-t', dest='task_num', type=int, default=0,
help='a way to separate different runs of the same parameter-set')
args = parser.parse_args()
params = config.get_config(args.param_set)
set_log_file(args.outname)
if args.file:
params['dataset_name'] = args.file
elif 'dataset_name' not in params:
params['dataset_name'] = default_dataset
run(0, **params)
print "finished"
with open(fname, 'w') as f:
writer = csv.writer(f, delimiter='\t')
writer.writerow(headers)
for i in idxs:
student = data['subject'][i]
outcome = 'correct' if data['correct'][i] == 2 else 'incorrect'
KCs = 'word' if single_skill else data.orig['skill'][i]
word_feats = skills[data['skill'][i]]
row = [student, outcome, KCs] + list(word_feats)
writer.writerow(row)
def to_fast(data, train_fname=None, valid_fname=None, fold=0, vector_length=150, single_skill=False):
train_fname = train_fname or 'FAST+deepkt_train{}.txt'.format(fold)
valid_fname = valid_fname or 'FAST+deepkt_test{}.txt'.format(fold)
valid_idx, train_idx = cv_split(data, cv_fold=fold, no_new_skill=True)
skills = gen_word_matrix(data.get_data('skill'), data['skill'].enum_pairs, vector_length=vector_length)
# create fast header
feat_headers = ['*features_{}'.format(i) for i in xrange(vector_length)]
headers = ['student', 'outcome', 'KCs'] + feat_headers
_to_fast_with_idxs(train_fname, headers, skills, data, train_idx, single_skill=single_skill)
_to_fast_with_idxs(valid_fname, headers, skills, data, valid_idx, single_skill=single_skill)
set_log_file('temp.log')
#for fold in xrange(14):
# to_fast(fold=fold, single_skill=False)
data = prepare_data(dataset_name='data/data5.gz', top_n=14)
to_fast(data, fold=0, single_skill=False)
def __enter__(self):
set_log_file(self.log_name)
def __enter__(self):
set_log_file(self.log_name)
with open('{}.model'.format(output_name), "wb") as f:
pickle.dump(model, f)
if __name__ == '__main__':
args = docopt(__doc__)
# load args
params = config.get_config(args['--param_set'])
err_filename = args['--err'] or gen_log_name()
out_filename = args['--out'] or "{}.model".format(err_filename)
if args['--quiet']:
log_filename = os.devnull
out_filename = os.devnull
print("Suppressing logging and output.")
set_log_file(err_filename)
task_num = int(args['--task_number'])
params['model'] = args['--model']
params['features'] = args['--feature']
params['conds'] = args['--cond']
params['dataset_name'] = args['--in']
params['data_mode'] = args['--data_mode']
params['output_name'] = out_filename
params['task_num'] = int(args['--task_number'])
if args['run']:
run(**params)
# TODO: figure out a good interface
else:
raise Exception("model type is not valid")
dataset = prepare_data(**kwargs)
train_idx, valid_idx = cv_split(dataset, percent=0.1, fold_index=task_num)
prepared_data = (dataset, train_idx, valid_idx)
model = SelectedModel(prepared_data, **kwargs)
model.train_full(**kwargs)
if __name__ == '__main__':
default_dataset = "raw_data/all_siegle.txt"
args = docopt(__doc__)
params = config.get_config(args['--param_set'])
log_filename = args['--out'] or gen_log_name()
if args['--quiet']:
log_filename = os.devnull
print("Not printing to log file.")
set_log_file(log_filename)
if args['--file']:
params['dataset_name'] = args['--file']
elif 'dataset_name' not in params:
params['dataset_name'] = default_dataset
params['conds'] = ['EyesClosed', 'EyesOpen']
run(0, **params)
print("Finished")