search

jehangiramjad / tslib

Time Series Forecasting and Imputation
48 stars 25 forks source link

[Q] Is normalization needed before running mRSC #3

Open JonathanTaws opened 5 years ago

JonathanTaws commented 5 years ago

Hi,

Thanks for making available your implementation of mRSC! Given that multidimensional timeseries can have different scales, is it needed/beneficial in theory to perform normalization before feeding in the data?

vishalmisra commented 4 years ago

Yes normalization always helps, it allows you to weigh different metrics appropriately.

JonathanTaws commented 4 years ago

@vishalmisra would there be a recommended way of doing normalization over the multivariate timeseries? Normalization can introduce biases in the data and the relationships the algorithm is trying to learn.