mxnet.gluon.contrib模块是否存在bug?


#1

用mxnet.gluon.contrib之rnn.Conv2DLSTMCell创建网络,训练后读取模型,用两种方法都报错:

  1. 用 gluon.SymbolBlock.imports()读取,错误如下:
    AssertionError: Parameter ‘data0’ is missing in file ‘f:/swap/result/firstlstm-0002.params’, which contains parameters: ‘conv0_weight’, ‘conv0_bias’, ‘conv1_weight’, …, ‘layernorm0_gamma’, ‘layernorm0_beta’, ‘dense0_weight’, ‘dense0_bias’. Please make sure source and target networks have the same prefix.
    2.用mx.model.load_checkpoint()读取,错误如下:
    ValueError: You created Module with Module(…, data_names=[‘data’]) but input with name ‘data’ is not found in symbol.list_arguments(). Did you mean one of:
    data0
    data1
    data2
    3.读取的文件确认是正确的,数据是训练用数据,确保格式是一致的。
    4.用两种办法读取不包含gluon.contrib模块的网络训练的结果,运行正常。
    所以怀疑可能存在bug。