Add example notebook for using aeon distances with sklearn clusterers #2511
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Thank you for contributing to
|
|
Dear maintainers, Clustering Overview |
hi, thanks for this, we will take a look |
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
- In the beginning, the details on how to use the sklearn estimators are not necessary. Please remove the "with metric=..."-stuff from the introduction and the TOC.
- The TOC does not render correctly
- There is no new content in the _Introduction_-section; can be removed.
- Please rename the section "Loading Data" to "Example Dataset"
Reply via ReviewNB
There was a problem hiding this comment.
Resolved conflicts and updated the notebook as per the review:
- Removed unnecessary "metric=..." details.
- Renamed "Loading Data" to "Example Dataset."
- Deleted the redundant "Introduction" section.
Please let me know if there are any additional changes required.
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
Why so complicated? You could just do
from aeon.datasets import load_unit_test X, y = load_unit_test(split="train") # [...]
Reply via ReviewNB
There was a problem hiding this comment.
Simplified dataset loading as per your feedback. Let me know if further changes are needed!
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
Please link to https://www.aeon-toolkit.org/en/stable/api_reference/distances.html for an overview of all aeon distances.
Reply via ReviewNB
There was a problem hiding this comment.
Added a link to the aeon distances API reference in the "Computing Distance Matrices with aeon" section as required
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
Please include:
AgglomerativeClusteringis, as the name suggests, an agglomerative approach that works by merging clusters bottom up.- Not all linkage methods can be used with a precomputed distance matrix.
single,complete,average, andweightedwork with aeon distances.
Reply via ReviewNB
There was a problem hiding this comment.
Just updated the "Hierarchical Clustering" section with an explanation of AgglomerativeClustering and the supported linkage methods for precomputed distance matrices, as required. Let me know if further changes are required
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
For all estimators: Please link to the corresponding scikit-learn documentation page.
Reply via ReviewNB
There was a problem hiding this comment.
Added links to the corresponding scikit-learn documentation pages for AgglomerativeClustering, DBSCAN, OPTICS, and SpectralClustering, as requred. Let me know if there are additional updates required
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
Line #7. inverse_distance_matrix = 1 /(distance_matrix + epsilon)
Unfortunately, this conversion approach does not preserve the properties of the distance distribution.
aeon "distances" (they are rather dissimilarities) are in the interval inverse_distance_matrix = 1 - (distance_matrix / distance_matrix.max())
Reply via ReviewNB
There was a problem hiding this comment.
Thanks, for the feedback. Just updated the distance-to-similarity conversion to normalize the distance matrix to [0, 1] and subtract from 1, as required. This ensures that the properties of the distance distribution are preserved. Let me know if further updates are required!
| @@ -0,0 +1,274 @@ | |||
| { | |||
There was a problem hiding this comment.
There was a problem hiding this comment.
Removed the references section as it was not cited in the notebook, as suggested.
SebastianSchmidl
added
documentation
Please refer to the links in my comment of the corresponding issue. |
- Removed "metric=..." details from TOC and introduction. - Renamed "Loading Data" to "Example Dataset." - Deleted redundant Introduction section.
"For a comprehensive overview of all available distance metrics in aeon, see the aeon distances API reference."
"AgglomerativeClustering is, as the name suggests, an agglomerative approach that works by merging clusters bottom-up." Clarified Supported Linkage Methods: Included the supported linkage methods (single, complete, average, weighted) for precomputed distance matrices.
…subtract from 1, ensuring proper preservation of distance distribution.
…istances.ipynb) in the Clustering Overview under Clustering Notebooks.
Thank you for the guidance! I have added a reference in the Please let me know if there are additional updates or adjustments required! |
Reference Issues/PRs
Fixes #1241
What does this implement/fix? Explain your changes.
This pull request introduces a new Jupyter Notebook: sklearn_clustering_with_aeon_distances.ipynb. The notebook demonstrates how to integrate aeon's distance metrics with scikit-learn clustering algorithms. It includes:
Hierarchical Clustering: Using AgglomerativeClustering with metric="precomputed".
Density-Based Clustering: Using DBSCAN and OPTICS with metric="precomputed".
Spectral Clustering: Using SpectralClustering with affinity="precomputed" and the inverse of the distance matrix as the similarity matrix.
This addition enhances the clustering documentation, showing how to combine aeon’s distance metrics with widely-used scikit-learn clusterers.
Does your contribution introduce a new dependency? If yes, which one?
No new dependencies introduced.
Any other comments?
The notebook has been tested locally, and all cells execute without errors.
A reference to this notebook has been added to the clustering section of the documentation.
PR checklist
For all contributions
For new estimators and functions
__maintainer__at the top of relevant files and want to be contacted regarding its maintenance. Unmaintained files may be removed. This is for the full file, and you should not add yourself if you are just making minor changes or do not want to help maintain its contents.For developers with write access