诡异的MxNet

1&＃xff0c;symbol命名方式&＃xff1a;
metric的名字和loss的名字要对应&＃xff0c;否则报错&＃xff1a;

Traceback (most recent call last):File "/home/user1/arcface-mtl/recognition/train_1118.py", line 390, in <module>main()File "/home/user1/arcface-mtl/recognition/train_1118.py", line 386, in maintrain_net(args)File "/home/user1/arcface-mtl/recognition/train_1118.py", line 378, in train_netepoch_end_callback&＃61;epoch_cb)File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/module/base_module.py", line 533, in fitself.update_metric(eval_metric, data_batch.label)File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/module/module.py", line 775, in update_metricself._exec_group.update_metric(eval_metric, labels, pre_sliced)File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/module/executor_group.py", line 640, in update_metriceval_metric.update_dict(labels_, preds)File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/metric.py", line 349, in update_dictmetric.update_dict(labels, preds)File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/metric.py", line 124, in update_dictpred &＃61; [pred[name] for name in self.output_names]File "/home/user1/miniconda3/lib/python3.7/site-packages/mxnet/metric.py", line 124, in <listcomp>pred &＃61; [pred[name] for name in self.output_names]
KeyError: &＃39;Id_output&＃39;


规则是 metric的名字比loss名字后面要多个_output

ce_lossRac &＃61; mx.symbol.MakeLoss(ce_lossRac, name&＃61;&＃39;lossRac&＃39;)eval_metric.add(Multi_Accuracy(num&＃61;4,output_names&＃61;[&＃39;xxx&＃39;, &＃39;xxx&＃39;, &＃39;xxx&＃39;, ,&＃39;lossRac_output&＃39;]))eval_metric.add(Multi_Loss(num&＃61;4,output_names&＃61;[&＃39;xxx&＃39;, &＃39;xxx&＃39;, &＃39;xxx&＃39;, &＃39;lossRac_output&＃39;]))