Performance of DistPred on Forecasting Dataset barely equivalent to MinusFormer Baseline

Anoise / DistPred

DistPred: A Distribution-Free Probabilistic Inference Method for Regression and Forecasting

29 stars 7 forks source link

Performance of DistPred on Forecasting Dataset barely equivalent to MinusFormer Baseline #2

Open iFe1er opened 1 month ago

iFe1er commented 1 month ago

I notice that the official code script is using Miniformer, but the paper are comparing performance with iTransformer, which is not a fair comparision.

Actually, the Performance of DistPred on Forecasting Dataset barely equivalent (some better, some are worse) to MinusFormer Baseline.

Evidence provided: Table from MinusFormer:

Table from DistPred:

Anoise commented 1 week ago

Later, we will add comparisons of the same backbone. We believe DistPred will still have good performance. We want to emphasize DistPred's uncertainty quantification ability rather than its predictive ability. Thank you for your attention.