Function build_downstream_solver

(cfg, dataset)

Source from the content-addressed store, hash-verified

95
96
97	def build_downstream_solver(cfg, dataset):
98	train_set, valid_set, test_set = dataset.split()
99	if comm.get_rank() == 0:
100	logger.warning(dataset)
101	logger.warning("#train: %d, #valid: %d, #test: %d" % (len(train_set), len(valid_set), len(test_set)))
102
103	if cfg.task['class'] == 'MultipleBinaryClassification':
104	cfg.task.task = [_ for _ in range(len(dataset.tasks))]
105	else:
106	cfg.task.task = dataset.tasks
107	task = core.Configurable.load_config_dict(cfg.task)
108	if not "lr_ratio" in cfg:
109	cfg.optimizer.params = task.parameters()
110	else:
111	cfg.optimizer.params = [
112	{'params': task.model.model.parameters(), 'lr': cfg.optimizer.lr * cfg.lr_ratio},
113	]
114	cfg.optimizer.params = task.parameters()
115	optimizer = core.Configurable.load_config_dict(cfg.optimizer)
116	solver = core.Engine(task, train_set, valid_set, test_set, optimizer, **cfg.engine)
117
118	if cfg.get("checkpoint") is not None:
119	solver.load(cfg.checkpoint)
120
121	if cfg.get("model_checkpoint") is not None:
122	if comm.get_rank() == 0:
123	logger.warning("Load checkpoint from %s" % cfg.model_checkpoint)
124	cfg.model_checkpoint = os.path.expanduser(cfg.model_checkpoint)
125	model_dict = torch.load(cfg.model_checkpoint, map_location=torch.device('cpu'))
126	task.model.load_state_dict(model_dict)
127
128	return solver
129
130
131	def build_pretrain_solver(cfg, dataset):

nothing calls this directly

no outgoing calls

no test coverage detected