MLWave
MLWave
Adding it to mapper with parameters feels cluttery: ``` mapper.map( n_cubes=4, perc_overlap=0.1, max_sample_cluster=1000, sub_n_cubes=2, sub_perc_overlap=0.2) ``` Instead, I am thinking of (but haven't fleshed out yet) something like: ``` mapper.map(cover=km.cover.MultiScaleCubical(n_cubes=[4,10],...
Thanks, I totally missed KeplerMapper having support for that! So subtle the implementation ;). So I made a 'km.cover.RecursiveCubicalCover()', and it works (kinda). The problem is the overlap, it quickly...
Use one of these in combination with: - a density estimator (Gauss density, ...) - outlier detector (NNeighbour distance, isolation forest, ...) - dimensionality reduction component (SVD, NMF, UMAP, ...)...
Cool, we can do that. Originally this was added, because some cluster algorithms in Scikit-learn exit with an error if you try clustering data that has a size less than...
As for 2. I can do two things. Easy: - create an extra function: .data_from_cluster_id(id) where id is an int (or maybe list of ints) with the cluster ID you...
```Python import km # Load digits data from sklearn import datasets data, labels = datasets.load_digits().data, datasets.load_digits().target # Initialize mapper = km.KeplerMapper(verbose=2) # Fit and transform data projected_data = mapper.fit_transform(data, projection=km.manifold.TSNE(random_state=1))...
> Does the current code assume palettes of a certain length? Yes. I think I could make the custom palette work and apply proper normalization. If we set a default...
KeplerMapper's API has not evolved since its first iteration. Back then I only clustered on the projection and it was more of a single-function-does-everything. Right now, I dont even know...
In my workflow I found it useful to call .fit_transform multiple times. I also like that behind the scenes it calls .fit_transform and that the output is a Numpy array....
> Let's stick with dicts and networkx objects then. Is it currently wired to accept networkx objects? Adding networkx as a requirement seems okay to me. I'll write a networkX...