def testone(): ##按默认执行
# get_data()
args = parser.parse_args()
np.random.seed(args.random_seed)
random.seed(args.random_seed)
args.do_model = 'mcls'
schema_labels, predict_data, predict_sents = process_data(args)
shiyan = """
mcls
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, predict_data, predict_sents, str(id))
def testone(id=1, do_model="mrc_relation"): ##按默认执行
# get_data()
args = parser.parse_args()
# get_submit_postprocess(args, id)
np.random.seed(args.random_seed)
random.seed(args.random_seed)
args.do_model = do_model
shiyan = """
mrc_relation
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
if (args.use_cross_validation):
id = cross_validation(args, id)
else:
schema_labels, predict_data, predict_sents = process_data(args, 4)
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, predict_data, predict_sents, str(id))
def bzsearch():
args = parser.parse_args()
# args.do_model = 'role'
schema_labels, predict_data, predict_sents = process_data(args)
# # 创建一个 LogWriter 对象 log_writer
# log_writer = LogWriter("./log", sync_cycle=10)
shiyan = """
######################################################################################################################################
trigger_batch_size gridsearch
######################################################################################################################################
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
id = 4 # str(datetime.now().strftime('%m%d%H%M'))
print(id)
for bz in [32, 16, 8]:
args.batch_size = bz
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, predict_data, predict_sents, str(id))
id += 1
def lrsearch():
args = parser.parse_args()
# args.do_model = 'role'
schema_labels, predict_data, predict_sents = process_data(args)
# # 创建一个 LogWriter 对象 log_writer
# log_writer = LogWriter("./log", sync_cycle=10)
shiyan = """
######################################################################################################################################
trigger_lrgridsearch
######################################################################################################################################
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
id = 1 # str(datetime.now().strftime('%m%d%H%M'))
print(id)
for lr in [3e-5, 1e-5, 1e-4]:
args.learning_rate = lr
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, predict_data, predict_sents, str(id))
id += 1
def my():
args = parser.parse_args()
args.do_model = 'role'
schema_labels = read_label('{}/entity2id.txt'.format(args.data_dir))
# # 创建一个 LogWriter 对象 log_writer
# log_writer = LogWriter("./log", sync_cycle=10)
shiyan = """
######################################################################################################################################
202,不复制，不考虑重叠
######################################################################################################################################
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
id = 5 # str(datetime.now().strftime('%m%d%H%M'))
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, str(id))
schema_labels, predict_data, predict_sents = process_data(args)
predict_by_model_path(args,args.checkpoint_dir, schema_labels, predict_data, predict_sents, id)
def lrepochsearch():
args = parser.parse_args()
args.do_model = 'role'
schema_labels = read_label('{}/entity2id.txt'.format(args.data_dir))
# # 创建一个 LogWriter 对象 log_writer
# log_writer = LogWriter("./log", sync_cycle=10)
shiyan = """
######################################################################################################################################
202,不复制，不考虑重叠,lrepochsearch
######################################################################################################################################
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
id = 6 # str(datetime.now().strftime('%m%d%H%M'))
for lr in [5e-5, 7e-5, 1e-4, 3e-4]: #[3e-5,1e-5,5e-6,1e-6]
args.learning_rate = lr
args.checkpoint_dir = 'models/' + args.do_model + str(id)
for epoch in range(1, 4):
args.num_epoch = epoch
one(args, schema_labels, str(id))
predict_by_model_path(args, args.checkpoint_dir, id)
id += 1
c.to_csv('./work/data.csv', header=None, index=False, sep='\t')
if __name__ == "__main__":
# get_data()
args = parser.parse_args()
np.random.seed(args.random_seed)
random.seed(args.random_seed)
args.do_model = 'role'
schema_labels, predict_data, predict_sents = process_data(args)
shiyan = """
######################################################################################################################################
trigger_batch_size lr gridsearch
######################################################################################################################################
"""
write_title('./work/log/' + args.do_model + '.txt', args, shiyan)
id = 1
args.checkpoint_dir = 'models/' + args.do_model + str(id)
one(args, schema_labels, predict_data, predict_sents, str(id))
# args = parser.parse_args()
# args.do_model = 'role'
# schema_labels, predict_data, predict_sents = process_data(args)
# # # 创建一个 LogWriter 对象 log_writer
# # log_writer = LogWriter("./log", sync_cycle=10)
#
# id = str(datetime.now().strftime('%m%d%H%M%S'))
# print(id)
#
# args.checkpoint_dir = 'models/trigger' + str(id)