def evaluate(args):
config = Config().from_file(args["config"])
from data.interface_lol import prepare_data_loader_4_lol
(train_loader, dev_loader, test_loader,
test_4d_loader) = prepare_data_loader_4_lol(args,
args["data_path"],
training=False)
from models.interface_lol import build_model_fn_4_lol
model_fn = build_model_fn_4_lol(args, config)
# for dev set
scf = loll.LoLLearner().build(
args, config,
os.path.join(ENV_PATH, "lol_{}_trained".format(args["dataset"])),
args["run_id"])
if args["run_dev_testing"]:
print("Development Set ...")
eval_spec = model_fn(ModeKeys.EVAL, dev_loader)
scf.evaluate(eval_spec)
if args['except_domain'] != "" and args["run_except_4d"]:
print("Test Set on 4 domains...")
eval_spec = model_fn(ModeKeys.EVAL, test_4d_loader)
scf.evaluate(eval_spec)
# for test set
print("Test Set ...")
eval_spec = model_fn(ModeKeys.TEST, test_loader)
scf.evaluate(eval_spec)
def evaluate(args):
config = Config().from_file(args["config"])
if args["dataset"] == "mwoz20":
from data.mwoz20.interface_meta_rl_lb import prepare_data_loader_4_meta_rl
(train_loader, dev_loader, test_loader,
test_4d_loader) = prepare_data_loader_4_meta_rl(args,
args["data_path"],
training=False)
from models.mwoz20.interface_meta_rl2 import build_model_fn_4_meta_rl
model_fn = build_model_fn_4_meta_rl(args, config)
# for dev set
scf = meta_rl2.MetaRLLearner().build(
args, config, os.path.join(ENV_PATH, "meta_rl2_trained"),
args["run_id"])
if args["run_dev_testing"]:
print("Development Set ...")
eval_spec = model_fn(ModeKeys.EVAL, dev_loader)
scf.evaluate(eval_spec)
if args['except_domain'] != "" and args["run_except_4d"]:
print("Test Set on 4 domains...")
eval_spec = model_fn(ModeKeys.EVAL, test_4d_loader)
scf.evaluate(eval_spec)
# for test set
print("Test Set ...")
eval_spec = model_fn(ModeKeys.TEST, test_loader)
scf.evaluate(eval_spec)
else:
raise Exception("Unimplemented")
def train(args):
config = Config().from_file(args["config"])
from data.interface_lol import prepare_data_loader_4_lol
(train_loader, dev_loader, test_loader,
test_4d_loader) = prepare_data_loader_4_lol(
args, args["data_path"], training=True)
from models.interface_lol_c2f import build_model_fn_4_lol
model_fn = build_model_fn_4_lol(args, config)
train_spec = model_fn(ModeKeys.TRAIN, train_loader)
eval_spec = model_fn(ModeKeys.EVAL, dev_loader)
scf = loll.LoLLearner().build(
args, config, os.path.join(
ENV_PATH, "lol_{}_trained".format(args["dataset"])), args["run_id"])
scf.train(train_spec, eval_spec)
def train(args):
config = Config().from_file(args["config"])
if args["dataset"] == "mwoz20":
from data.mwoz20.interface_meta_rl_lb import prepare_data_loader_4_meta_rl
(train_loader, dev_loader, test_loader,
test_4d_loader) = prepare_data_loader_4_meta_rl(args,
args["data_path"],
training=True)
from models.mwoz20.interface_meta_rl2 import build_model_fn_4_meta_rl
model_fn = build_model_fn_4_meta_rl(args, config)
train_spec = model_fn(ModeKeys.TRAIN, train_loader)
eval_spec = model_fn(ModeKeys.EVAL, dev_loader)
scf = meta_rl2.MetaRLLearner().build(
args, config, os.path.join(ENV_PATH, "meta_rl2_trained"),
args["run_id"])
scf.train(train_spec, eval_spec)
else:
raise Exception("Unimplemented")