Support for different sampling rate across different channels/variables?

[findalexli]

As the topic suggests, I am highly interested to see if this library can support multivariable time series with channels that have different sampling rate and thus different length within a sample For instance, From this paper, they implemented channel-independent path which are all concated at the end before MLP layers. The legacy tensorflow code is provided as follows: Can We Ditch Feature Engineering? End-to-End DeepLearning for Affect Recognition from Physiological Sensor Data https://github.com/Emognition/dl-4-tsc