cure-lab / SCINet

The GitHub repository for the paper: “Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction“. (NeurIPS 2022)
Apache License 2.0
617 stars 127 forks source link

Is it normal for the MAPE value to be too large? #75

Open yaya-eng opened 9 months ago

yaya-eng commented 9 months ago

Hello, I made a prediction on dataset ETTH1 using your code, and the results of mae and mse are similar to those in the paper. I have two questions to ask.

  1. The result of MAPE calculation in multivariate prediction is approximately 8-9 (but ideally it should be a number between 0 and 1). Later, I calculated the map e of the OT variable separately as 0.128, but the map e values of other variables are relatively large. Is this reasonable?
  2. It was found that both MAE and MSE were calculated using normalized data. Should we use the data before normalization for calculation? 运行图片