Implementation of NCH (Nearest Convex Hull) classifier #253
Conversation
It gives an error: AttributeError: Pipeline has none of the following attributes: decision_function.
Added an example that runs on a small number of test samples, so that we can get results quicker.
- limite size of training set - change to slsqp optimizer
Merge branch 'main' of https://github.com/toncho11/pyRiemann-qiskit-rep
Added second parameter that specifies the number of hulls.
Added support for transform(). Added a new pipeline [NCH+LDA]
There was a problem hiding this comment.
Good job @toncho11!
I see you introduced the n_hulls and n_samples_per_hull parameters.
We could do some hyperparmetrization in a follow-up PR.
I see that the signature of the distance function differs a little bit from pyRiemann, and this is why you could not reuse the whole suggestion from @qbarthelemy in the _predict_distances method, but you introduced the transform.
Some minor comments. Questions:
- Do we keep the classifly_P300_nch.py? (I think it could be handfull for the next iterations)
- Do we want to do a short test with quantum=True before merging?
Improvements requested by GC.
|
I think the PR is ready for merge. |
Set of SPD matrices. Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Added new lines to before Parameters Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
[y == c, :, :] => [y == c] Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
NearestConvexHull text change Co-authored-by: Quentin Barthélemy <q.barthelemy@gmail.com>
Changes about the global optimizer so, that it is more evident that a global one is used.
Added support for both "min-hull" and "random-hull" using the constructor parameter "hull-type".
| predictions = [ | ||
| self.classes_[min(range(len(values)), key=values.__getitem__)] | ||
| for values in dist | ||
| ] |
There was a problem hiding this comment.
What is the difference with predictions = self.classes_[dist.argmin(axis=1)] ?
| for test_sample in X: | ||
| dist_sample = self._process_sample(test_sample) | ||
| dist.append(dist_sample) | ||
|
|
There was a problem hiding this comment.
You should add dist = np.concatenate(dist, axis=1) here, in order to transform list into array.
| distances_to_covs = [ | ||
| distance_logeuclid(test_sample, cov) | ||
| for cov in self.matrices_per_class_[c] | ||
| ] |
There was a problem hiding this comment.
Using from pyriemann.utils.distance import distance, you can simplify code this way
distances_to_covs = distance(self.matrices_per_class_[c], test_sample, metric="logeuclid")[:, 0]
Then, you can compute np.argsort(distances_to_covs) instead of np.argsort(np.array(distances_to_covs)).
QuanticNCH uses NearestConvexHull classifier (non quantum version)
Resolves #246
Resolves #254