Download specific UCR_UEA dataset about tslearn HOT 4 CLOSED

Hi, and thanks for your feedback!

Could you split this in two separate issues, since I think there are two things discussed here.

Let's keep this issue for the first problem you point out: downloading separate datasets.
At the moment, what the code does is download an archive from www.timeseriesclassification.com, unzip it locally and access datasets locally.
In order to do what you suggest, we should host the datasets by ourselves (we should ask UCR/UEA for permission, I guess + we should decide on where to host the data, and how to pay for the hosting service :) ).

Maybe that could be a good solution (if UCR/UEA guys are OK, once again), but I have no idea how large a single dataset in .npz format could be.

Finally, for testing purposes, there exists a CachedDatasets class, I don't know if you've seen it: http://tslearn.readthedocs.io/en/latest/gen_modules/datasets/tslearn.datasets.CachedDatasets.html#tslearn.datasets.CachedDatasets (at the moment, only Trace is cached).

from tslearn.

Thanks for your response @rtavenar !

Could you split this in two separate issues, since I think there are two things discussed here.

Done, moved the second part to a separate issue #23

In order to do what you suggest, we should host the datasets by ourselves (we should ask UCR/UEA for permission, I guess + we should decide on where to host the data, and how to pay for the hosting service :) ).

Well generally it's a bit of a pain to host datasets. I know that scikit-learn made a copy of datasets it depends on at https://figshare.com/ (scikit-learn/scikit-learn#7425)

But in this issue I was just thinking that as far as I saw www.timeseriesclassification.com does provide a fairly straightforward download link for each dataset, e.g.

http://www.timeseriesclassification.com/Downloads/DiatomSizeReduction.zip

which is

"http://www.timeseriesclassification.com/Downloads/%s.zip" % dataset_name

so I guess it's possible to only download those datasets that are explicitly loaded (then cache them as it is done currently). I'have not checked whether this generalizes to all datasets though...

from tslearn.

Oh, great! I didn't know that.

Then we should have a way to deal with it, definitely.

from tslearn.

This behaviour is now implemented in tslearn (>= 0.1.12)

from tslearn.