Extraction of Shapelets and return as a list about wildboar HOT 10 CLOSED

Hi,

You can find the shapelets in the tree attribute. forest.estimators_[0].tree.features give a list of the shapelets of the first tree in the ensemble. Check the _tree.py file for details. The tree has the same structure as a tree in sklearn so examples for extracting features of such trees can be adapted.

from wildboar.

Thanks, I've managed to retrieve shapelets.
I have another question: the forest.estimators_[0].tree.feateures list contains several values.

For example:
features[0] = (1, array([....]))
features[1] = None
features[2] = (1, array([....]))
features[3] = None
features[4] = None
etc...

It seems that some elements are None, while other are in the form (number, array). I think that maybe the array is the shapelet and the number is the target class.

Why are some elements None? Why does the number of elements in the list differ from a tree to another?

Thank you again!

from wildboar.

If the value is None the node index refers to a leaf node. The first element of the tuple refers to the dimension of the input data and the second is the shapelet.

from wildboar.

The number of features depend on the depth of the tree. Not all trees are equally deep.

from wildboar.

Here is an example where all branches are visited: https://github.com/isaksamsten/wildboar/blob/9348d46422bf9665b5aa67f17aa397191764e85b/src/wildboar/explain/counterfactual/_sf.py#L71

from wildboar.

Hi @isaksamsten ,
I stumbled upon this thread while looking for similar functionality...in my own usage, I found that I was able to access the shapelets as (assuming the classifier object is called clf and assuming the first tree at index 0):

clf.estimators_[0].estimator.tree_.shapelet

Rather than the way you described above. Is this way correct or am I missing something?
Thanks!

from wildboar.

Hi!

You are not missing anything. For version 1.1, the shapelet attribute is renamed features.

from wildboar.

Hello!
Ah, got it. v1.1 is seemingly not on pip yet, so I'll just make a note to update my code when it comes out. Thanks for the clarification and the amazing work!

from wildboar.

Not yet. The update is rather massive and I want to make sure that there aren’t too many regressions. (I will try to write a migration guide too 😱)

from wildboar.

Haha, that's ok. Thank you once again for the amazing work!

from wildboar.