Documentation Inconsistency

[BradKML]

Currently flameplot only provides a "fig" as a tuple of fig (plot) and ax
https://github.com/erdogant/flameplot/blob/master/flameplot/flameplot.py#L161

But the tuple contains the real fig (plot) and ax
https://github.com/erdogant/scatterd/blob/master/scatterd/scatterd.py#L156

For purposes of saving graphs, this could be fixed.

For testing (given a proper X):

!pip install umap-learn trimap pacmap
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from umap import UMAP # https://github.com/lmcinnes/umap
from trimap import TRIMAP # https://github.com/eamid/trimap
from pacmap import PaCMAP # https://github.com/YingfanWang/PaCMAP

k = 3
mapper = [TSNE(n_components=k), 
  # UMAP(n_components=k),
  # TRIMAP(n_dims=k),
  PaCMAP(n_components=k)
]

from flameplot import scatter, compare, plot

pca_standard = PCA().fit_transform(X)

for i in mapper:
  title = type(i).__name__ # name of the object class
  transformed = i.fit_transform(X)
  fig, _ = scatter(transformed[:,0], transformed[:,1], labels=y['country'], s=75, title=title)
  fig.savefig(title+'_2D.png')
  scores = compare(pca_standard, transformed, n_steps=5)
  fig, _ = plot(scores, xlabel='PCA', ylabel=title)
  fig.savefig(title+'_factors.png')

[erdogant]

Is your suggestion to only return fig and not ax?

[BradKML]

Current edit might makes things clearer, it might be good to always return fig and ax, and have the documentation match the standard output.

!pip install umap-learn trimap pacmap
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from umap import UMAP # https://github.com/lmcinnes/umap
from trimap import TRIMAP # https://github.com/eamid/trimap
from pacmap import PaCMAP # https://github.com/YingfanWang/PaCMAP

k = 3
mapper = [TSNE(n_components=k), UMAP(n_components=k), TRIMAP(n_dims=k), PaCMAP(n_components=k)]

from  flameplot import scatter, compare, plot

pca_standard = PCA().fit_transform(X.to_numpy())

def scatter_maker(transformed, title, labels=y['country']):
  fig1, _ = scatter(transformed[:,0], transformed[:,1], labels=y, s=75, title=title)
  # documentation claimed there is only `(fig)` and not `(fig, ax)`
  # https://github.com/erdogant/scatterd/blob/master/scatterd/scatterd.py#L156
  # https://github.com/erdogant/flameplot/blob/master/flameplot/flameplot.py#L161
  fig1.savefig(title+'_2D.png')
  pass

def compare_maker(transformed, title, pca_standard=pca_standard):
  scores = compare(pca_standard, transformed, n_steps=5)
  fig2 = plot(scores, xlabel='PCA', ylabel=title) # not follow the `(fig, ax)` design pattern
  # https://github.com/erdogant/imagesc/blob/master/imagesc/imagesc.py#L243
  # https://github.com/erdogant/flameplot/blob/master/flameplot/flameplot.py#L138
  fig2.savefig(title+'_factors.png')
  pass

for i in mapper:
  title = type(i).__name__ # name of the object class
  transformed = i.fit_transform(X.to_numpy())
  # to_numpy is needed for TriMAP and PaCMAP
  scatter_maker(transformed, title, y)
  compare_maker(transformed, title, pca_standard=pca_standard)

[erdogant]

Agree. The output is now (fig, ax) and is also changed in the documentation pages.

Update to the latest version with:

pip install -U flameplot