Mascot UMAP

mascot_umap.py

from PIL import image
import pandas as pd
import numpy as np

empty_pixel = np.array([255, 255, 255, 0]).reshape(1, 1, -1)

# !wget https://github.com/theislab/scanpy/raw/master/docs/_static/img/Scanpy_Logo_RGB.png
im = Image.open("./Scanpy_Logo_RGB.png")
a = np.array(im)

# Make dataframe of pixels
df = pd.DataFrame(
    dict(
        zip(
            ("y_coord", "x_coord"),
            np.where(np.logical_and.reduce(~np.equal(a, empty_pixel), axis=2))
        )
    )
)
# Translate pixel coordinates from cartesian to euclidean
df["y"] = np.abs(df["y_coord"] - df["y_coord"].max())
df["x"] = df["x_coord"]

# Remove the text
sub = df[~((df["y"] < 400) & (df["x"] < 1500))]

# Retrieving rgba values:
colors = a[sub["y_coord"], sub["x_coord"], :]
for i, c in enumerate("rgba"):
    sub[c] = colors[:, i]

import umap
import umap.plot

u = umap.UMAP()
u.fit(sub.drop(columns=["x_coord", "y_coord"]))

# But I couldn't figure out how to use the rgba info for a color map, so I'll color by:
sub["c"] = sub["x"] - sub["y"]
sub["c"] -= sub["c"].min()
sub["c"] /= sub["c"].max()

umap.plot.points(u, values=sub["c"], cmap="magma_r", width=3000, height=3000)

# Hexbin

import hvplot.pandas

sub["inv_c"] = 1 - sub["c"]
(
    sub
    .hvplot.hexbin("x", "y", C="inv_c", reduce_function=np.sum, cmap="magma", logz=True, xaxis=None, yaxis=None, colorbar=False, width=600, height=350)
    .opts(gridsize=45)
)