Tables for paper "Time Series Forecasting Using Recurrent Neural Networks Based on Recurrent Sigmoid Piecewise Linear Neurons"

This document contains tables for the paper "Time Series Forecasting Using Recurrent Neural Networks Based on Recurrent Sigmoid Piecewise Linear Neurons". Description of each table in order (Table 1 to Table 3):RMSSE on 30 test time series (taken from the M5 competition dataset) of "C1-type networks" with different neuron types, specifically: LSTM, GRU, recurrent sigmoid piecewise linear (RSPL) neuron and recurrent sigmoid piecewise linear neuron without context normalization (RSPLWCN). C1-type network in the context of this paper means: a small neural network with 2 inputs, 1 recurrent layer with a 4-dimensional context vector.RMSSE on 30 test time series of "C2-type networks" for the same neuron types as above. C2-type network in the context of this paper means: a relatively big neural network with 4 inputs, 3 recurrent layers with an 8-dimensional context vector.Results from 2 previous tables, aggregated over all test time series.

本文件包含论文《基于递归sigmoid分段线性神经元的递归神经网络时间序列预测》的相关表格。各表格（表1至表3）的依次说明如下：不同神经元类型的“C1型网络”在30个测试时间序列（取自M5竞赛数据集）上的均方根缩放误差（RMSSE），具体神经元类型包括：长短期记忆网络（LSTM）、门控循环单元（GRU）、递归sigmoid分段线性（RSPL）神经元以及无上下文归一化的递归sigmoid分段线性（RSPLWCN）神经元。本文语境下的C1型网络指：具有2个输入、1个递归层（含4维上下文向量）的小型神经网络。上述相同神经元类型的“C2型网络”在30个测试时间序列上的RMSSE。本文语境下的C2型网络指：具有4个输入、3个递归层（含8维上下文向量）的相对大型神经网络。前两张表格结果在所有测试时间序列上的聚合值。