An offical implementation of PatchTST: "A Time Series is Worth 64 Words: Long-term Forecasting with Transformers." (ICLR 2023) https://arxiv.org/abs/2211.14730
Hi, the length of the unknown data I want to predict is 4380, how should I make a prediction for such a long length, should I train a model structure and make a prediction, and then feed the predicted data into the model again, one predicting the data for the next time period, or should I incorporate the predicted data into the dataset for a new training? I tried the first option but after scrolling forward a few times the error accumulation is very large and there is a problem with the error distribution(the correct error distribution should be double hump, while my error distribution is similar to a pta distribution), how should I go about this please, thank you for your answer!
Hi, the length of the unknown data I want to predict is 4380, how should I make a prediction for such a long length, should I train a model structure and make a prediction, and then feed the predicted data into the model again, one predicting the data for the next time period, or should I incorporate the predicted data into the dataset for a new training? I tried the first option but after scrolling forward a few times the error accumulation is very large and there is a problem with the error distribution(the correct error distribution should be double hump, while my error distribution is similar to a pta distribution), how should I go about this please, thank you for your answer!