def network(self):
model = get_model(self.config)
self.input_data = model.create_feeds()
self.metrics = model.net(self.input_data)
self.inference_target_var = model.inference_target_var
logger.info("cpu_num: {}".format(os.getenv("CPU_NUM")))
model.create_optimizer(get_strategy(self.config))
def init_network(self):
model = get_model(self.config)
self.input_data = model.create_feeds()
self.metrics = model.net(self.input_data)
self.inference_feed_vars = model.inference_feed_vars
self.inference_target_var = model.inference_target_var
if hasattr(model, "all_vars"):
with open("all_vars.txt", 'w+') as f:
f.write('\n'.join([var.name for var in model.all_vars]))
if config.get("runner.need_prune", False):
# DSSM prune net
self.inference_feed_vars = self.model.prune_feed_vars
self.inference_target_var = self.model.prune_target_var
if config.get("runner.need_train_dump", False):
self.train_dump_fields = model.train_dump_fields if hasattr(
self.model, "train_dump_fields") else []
self.train_dump_params = model.train_dump_params if hasattr(
self.model, "train_dump_params") else []
if config.get("runner.need_infer_dump", False):
self.infer_dump_fields = model.infer_dump_fields if hasattr(
model, "infer_dump_fields") else []
self.config['stat_var_names'] = model.thread_stat_var_names
self.metric_list = model.metric_list
self.metric_types = model.metric_types
logger.info("cpu_num: {}".format(os.getenv("CPU_NUM")))
model.create_optimizer(get_strategy(self.config))
def network(self):
self.model = get_model(self.config)
self.input_data = self.model.create_feeds()
self.metrics = self.model.net(self.input_data)
logger.info("cpu_num: {}".format(os.getenv("CPU_NUM")))
thread_stat_var_names = [
self.model.auc_stat_list[2].name, self.model.auc_stat_list[3].name
]
thread_stat_var_names += [i.name for i in self.model.metric_list]
thread_stat_var_names = list(set(thread_stat_var_names))
self.config['stat_var_names'] = thread_stat_var_names
self.metric_list = list(self.model.auc_stat_list) + list(
self.model.metric_list)
self.metric_types = ["int64"] * len(self.model.auc_stat_list) + [
"float32"
] * len(self.model.metric_list)
self.model.create_optimizer(get_strategy(self.config))