[ENH] Shapelet Transform improvements

[TonyBagnall]

the main thing here is that the conversion from numpy to dataframe is removed. All that was happening was the numpy was getting put into a dataframe in the transform, then converted back to numpy in classifier. The base transformer does this conversion automatically, unless you set this argument in the constructor

super(RandomShapeletTransform, self).init(_output_convert=False)

other changes:

1. Docstrings tidied up, dimensions changed to channels, all reference to DataFrames removed
2. Helper classes for full ShapeletTransform made private
3. Remove pointless tags

to do in a later PR
Order of classes in file changed to put RandomShapeleTransform first

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[GuzalBulatova]

I've got a couple of comments and a general suggestion to break down the big PRs into smaller ones where possible.
Overall this is absolutely fantastic work! Definitely a significant improvement. The code flow is easy to read and I really appreciate the comprehensive documentation.

[MatthewMiddlehurst]

I agree that it is extremely difficult to tell what has actually changed without going through the full file (both pre and post change). Maybe moving them around should be a separate thing.

[TonyBagnall]

thanks for the feedback @GuzalBulatova, really helpful. I did wonder about the complexity of reviewing when moving classes, just wanted to reprioritise since the ShapeletTransformClassifier is there for legacy only, but it is purely cosmetic, I will shift them back. I have a more systemic issues about the tests for which I will raise an issue

[TonyBagnall]

I agree that it is extremely difficult to tell what has actually changed without going through the full file (both pre and post change). Maybe moving them around should be a separate thing.

agreed, I will shift order on its own PR, hopefully more readable now

[TonyBagnall]

Im working on a notebook revamp for classification on a different branch, but I think as I am here with STC I will write a shapelet notebook, which make take a while

[TonyBagnall]

I have added a notebook on using the transform. I want to put more details in, but I need to work out how it interacts with the classification notebooks I'm changing on #123 so would like to leave it at this for now.

[MatthewMiddlehurst]

I'm fine with the changes, but the shapelet notebook may conflict with the previous notebook PR. Just not getting picked up because it's in a different dir.

made the changes, hope its ok to dismiss