Infer feature extraction settings for tsfresh

MaxBenChrist / tspreprocess

A Python package to preprocess time series

MIT License

60 stars 10 forks source link

Infer feature extraction settings for tsfresh #4

Open MaxBenChrist opened 7 years ago

MaxBenChrist commented 7 years ago

We could use tspreprocess to infer good settings for kind_to_fc_parameters.

For example

if all time series have same length, we do not need to calculate the length feature
for autocorrelation, we could use all lags up to lets say one tenth of the maximal length, so maxlag = max(len(x))/10
if the time series are discrete, maybe drop the trend features
...

This would allow to use tsfresh more efficiently

nils-braun commented 7 years ago

Hm yes - you are right in principle - however I do not know if we would save that much time here. I think the main feature of this package would be the denoising or the resampling - this is really where people can get the much out of it.