# ML for time series

## Topics of interest (unsorted list)

* preprocessing
* incl. representations for time series (PAA, SAX, etc.)
* classification / regression
* nearest neighbor based (dtw & variants)
* kernel based (gak)
* BOSS & co
* Shapelets
* ConvNets
* a word on ensembles?
* a word on early classification?
* clustering (incl. barycenter computation)
* kmeans with DTW (DBA) / softDTW
* KShape
* forecasting
* auto-regressive models
* hidden markov models?
* recurrent models (RNN, LSTM, LSTM à la Graves, ODE-RNN)
* GANs
* representation learning?
* are there methods that are not based on generative models?
* indexing/retrieval?

For each method, make sure the following is clear:
* [ ] works or not on multidim time series
* [ ] works or not on variable length time series
* [ ] impacted or not by irregularly sampled time series

## Potential people involved

* Romain Tavenard
* Johann Faouzi ? (classification)
* François Painblanc + encadrants ? (generative models)
* Simon Malinowski ? (HMMs & retrieval/indexing)
* Yann Soullard ? (HMMs & neural nets)
* Germain Forestier & Hassan Ismail Fawaz ? (ConvNets for classification / regression)

## Toolkits used

* `tslearn`
* `pyts` for classification methods
* `pytorch` for Neural Nets

## Tools to write the book

* https://github.com/jupyter/jupyter-book ?