search

gzerveas / mvts_transformer

Multivariate Time Series Transformer, public version
MIT License
718 stars 169 forks source link

Hyperparameters and accuracy for AtrialFibrillation dataset. #52

Open letsGoBharat opened 1 year ago

letsGoBharat commented 1 year ago

Hello,

I am attempting to perform pre-training and fine-tuning on the AtrialFibrillation dataset, but I am unable to locate the hyperparameters and corresponding performance metrics in the relevant paper. Can you please provide me with this information, if available?

letsGoBharat commented 1 year ago

Is it possible to get some help with this? @gzerveas

gzerveas commented 1 year ago

Hi, this dataset was not included in the evaluation of the paper. If it is in the time series regression/classification archive, you can download it and put it in a directory as the README indicates, and the class TSRegressionArchive(BaseData): in the data.py should be able to handle it. However, I can't really tell what would be good hyperparameters without having worked with this dataset. I suggest looking at the paper's table with default hyperparamters as a starting point, and experiment a bit (have a look at other datasets' hyperparameters to see which ones are the ones that most often affect performance). You may also consider using a sub-sample factor if the time series are very long or sampled with a high temporal resolution.