Created by: ARDUJS

环境

• paddlepaddle 1.7.2
• python 3.7

问题

如何将在动态图上写CRF?

这是在静态图上编写CRF的代码

       res = []
       ## 标签1的发射状态特征权重
       self.emission_1 = fluid.layers.fc(size=self.num_classes,input=unpad_feature,param_attr=fluid.ParamAttr(
                   initializer=fluid.initializer.Uniform(low=-0.1, high=0.1),
                   regularizer=fluid.regularizer.L2DecayRegularizer(regularization_coeff=1e-4)))
       size = self.emission_1.shape[1]
       ## 标签1的转移特征权重
       w = fluid.layers.create_parameter(
               shape=[size + 2, size], dtype=self.emission_1.dtype, name='crfw_1')
       ## 预测标签1
       self.ret_infers_1 = fluid.layers.crf_decoding(
               input=self.emission_1, param_attr=fluid.ParamAttr(name='crfw_1'))
       ret_infers = fluid.layers.assign(self.ret_infers_1)
       res.append(ret_infers)

我应该如何将其修改为动态图上的代码