Input/Output

Comparison of two embeddings.

Description

Quantification of local similarity across two maps or embeddings, such as PCA and t-SNE. To compare the embedding of samples in two different maps using a scale dependent similarity measure. For a pair of maps X and Y, we compare the sets of the, respectively, kx and ky nearest neighbours of each sample.

param mapX:

Mapping of first embedding.

type mapX:

numpy array

param data2:

Mapping of second embedding.

type data2:

numpy array

param nn:

number of neirest neighbor to compare between the two maps. This can be set based on the smalles class size or the aveage class size. The default is 250.

type nn:

integer, optional

param n_steps:

The number of evaluation steps until reaching nn, optional. If higher, the resolution becomes lower and vice versa. The default is 5.

type n_steps:

integer

param verbose:

print messages. The default is 3.

type verbose:

integer, optional

returns:

• scores : array with the scores across various nearest neighbors (nn).

• nn : nearest neighbors

• n_steps : The number of evaluation steps until reaching nn.

rtype:

dict()

Examples

>>> # Load data
>>> X, y = KNRscore.import_example()
>>>
>>> # Compute embeddings
>>> embed_pca = decomposition.TruncatedSVD(n_components=50).fit_transform(X)
>>> embed_tsne = manifold.TSNE(n_components=2, init='pca').fit_transform(X)
>>>
>>> # Compare PCA vs. tSNE
>>> scores = KNRscore.compare(embed_pca, embed_tsne, n_steps=10)
>>>
>>> # plot PCA vs. tSNE
>>> fig, ax = knrs.scatter(embed_tsne[:, 0], embed_tsne[:, 1], labels=y, cmap='Set1', title='tSNE Scatter Plot')
>>> fig, ax = knrs.scatter(embed_pca[:, 0], embed_pca[:, 1], labels=y, cmap='Set1', title='PCA Scatter Plot')
>>>

References

• Blog: https://towardsdatascience.com/the-similarity-between-t-sne-umap-pca-and-other-mappings-c6453b80f303

• Github: https://github.com/erdogant/KNRscore

• Documentation: https://erdogant.github.io/KNRscore/

Plot

Make plot.

param out:

output of the compare() function.

type out:

dict

param cmap:

colormap. The default is ‘jet’.

type cmap:

string, optional

param xlabel:

Label for x-axis. The default is None.

type xlabel:

str, optional

param ylabel:

Label for y-axis. The default is None.

type ylabel:

str, optional

param reverse_cmap:

Reverse the colormap. The default is False.

type reverse_cmap:

bool, optional

returns:

Figure and axes objects.

rtype:

fig, ax

Examples

>>> # Load data
>>> X, y = KNRscore.import_example()
>>>
>>> # Compute embeddings
>>> embed_pca = decomposition.TruncatedSVD(n_components=50).fit_transform(X)
>>> embed_tsne = manifold.TSNE(n_components=2, init='pca').fit_transform(X)
>>>
>>> # Create comparison scores
>>> scores = KNRscore.compare(embed_pca, embed_tsne)
>>>
>>> # Create plot with custom labels
>>> fig, ax = KNRscore.plot(scores, cmap='viridis', xlabel='PCA', ylabel='tSNE')
>>>
>>> # Create plot with reversed colormap
>>> fig, ax = KNRscore.plot(scores, cmap='viridis', reverse_cmap=True)
>>>

---

---