The Influence of Embedding Size and Hidden Units Parameters Changes on the Sequence2Sequence Model
Zhengfang He, Mingbo Pan, Yikai Wang, Gang Xu, Weibin Su, Chunmei Shi · 2022
The parameters of the neural network are very important to the model, and the changes of the parameters have a greater impact on the accuracy of the neural network. This paper takes the Sequence2Sequence (Seq2Seq) model as an example, in which Encoder and Decoder both use basic RNN models to research the impact of parameter changes on neural network performance. The EmbeddingSize and Hidden Units are two important parameters of the Seq2Seq model. This paper uses parameter combinations to build different models based on the two important parameters. And using self-made Sequential-Reverse data sets to experiment on these models. Then visualize the experimental results in three dimensions and analyze the influence of these two parameters on the Seq2Seq model. The results of this paper can provide researchers with some suggestions when they build neural networks.