Tamanho hexadecimal em hexbins matplotlib com base na densidade de pontos próximos

Eu tenho o seguinte código que produz a figura a seguir

import numpy as np
np.random.seed(3)
import pandas as pd
import matplotlib.pyplot as plt
df = pd.DataFrame()
df['X'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))
df['Y'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))

df['Bin'] = df.apply(lambda row: .1 if row['X'] < 30 and row['Y'] < 30 else .9, axis=1)

fig, ax = plt.subplots(figsize=(10,10))
plt.scatter(df['X'], df['Y'])

Eu gráfico os dados usando hexbins, como observado abaixo

from matplotlib import cm

fig, ax = plt.subplots(figsize=(10,10))
hexbin = ax.hexbin(df['X'], df['Y'], C=df['Bin'], gridsize=20, cmap= cm.get_cmap('RdYlBu_r'),edgecolors='black')
plt.show()

Gostaria de alterar o tamanho dos hexágonos com base na densidade dos pontos plotados na área que um hexágono cobre. Por exemplo, os hexágonos no canto inferior esquerdo (onde os pontos são compactos) serão maiores que os hexágonos em qualquer outro lugar (onde os pontos são escassos). Existe uma maneira de fazer isso?

Edit: Eu tentei esta solução , mas não consigo descobrir como colorir os hexágonos com base em df ['Bin'], ou como definir o tamanho hexadecimal mínimo e máximo.

from matplotlib.collections import PatchCollection
from matplotlib.path import Path
from matplotlib.patches import PathPatch
fig, ax = plt.subplots(figsize=(10,10))
hexbin = ax.hexbin(df['X'], df['Y'], C=df['Bins'], gridsize=20, cmap= cm.get_cmap('RdYlBu_r'),edgecolors='black')
def sized_hexbin(ax,hc):
    offsets = hc.get_offsets()
    orgpath = hc.get_paths()[0]
    verts = orgpath.vertices
    values = hc.get_array()
    ma = values.max()
    patches = []
    for offset,val in zip(offsets,values):
        v1 = verts*val/ma+offset
        path = Path(v1, orgpath.codes)
        patch = PathPatch(path)
        patches.append(patch)

    pc = PatchCollection(patches, cmap=cm.get_cmap('RdYlBu_r'), edgecolors='black')
    pc.set_array(values)
    ax.add_collection(pc)
    hc.remove()

sized_hexbin(ax,hexbin)
plt.show()

Você pode gastar algum tempo na compreensão do mapeamento de cores.

    import numpy as np
    np.random.seed(3)
    import pandas as pd
    import matplotlib.pyplot as plt
    from matplotlib.collections import PatchCollection
    from matplotlib.path import Path
    from matplotlib.patches import PathPatch
    df = pd.DataFrame()
    df['X'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))
    df['Y'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))

    df['Bin'] = df.apply(lambda row: .1 if row['X'] < 30 and row['Y'] < 30 else .9, axis=1)

    #fig, ((ax1, ax2)) = plt.subplots(1, 2, sharex=True, sharey=True)
    ax1 = plt.scatter(df['X'], df['Y'])

    fig,ax2 = plt.subplots(figsize=(10,10))
    hexbin = ax2.hexbin(df['X'], df['Y'], C=df['Bin'], gridsize=20,edgecolors='black',cmap= 'RdBu', reduce_C_function=np.bincount) #**

    def sized_hexbin(ax,hc):
        offsets = hc.get_offsets()
        orgpath = hc.get_paths()[0]
        verts = orgpath.vertices
        values = hc.get_array()
        ma = values.max()
        patches = []
        for offset,val in zip(offsets,values):
            v1 = verts*val/ma + offset
            path = Path(v1, orgpath.codes)
            patch = PathPatch(path)
            patches.append(patch)

        pc = PatchCollection(patches, cmap= 'RdBu', edgecolors='black')
        pc.set_array(values)

        ax.add_collection(pc)

        hc.remove()

    sized_hexbin(ax2,hexbin)
    cb = plt.colorbar(hexbin, ax=ax2)

    plt.show()

To plot the chart based on df['bins'] values - 

Need to change the reduce_C_function in #** marked line -

    hexbin = ax2.hexbin(df['X'], df['Y'], C=df['Bin'], gridsize=20,edgecolors='black',cmap= 'RdBu', reduce_C_function=np.sum)

[![enter image description here][2]][2]


  [1]: https://i.stack.imgur.com/kv0U4.png
  [2]: https://i.stack.imgur.com/mb0gD.png

# Another variation of the chart :

# Where size is based on count of points in the bins and color is based on values of the df['bin']./ Also added if condition to control minimum hexbin size.


import numpy as np
np.random.seed(3)
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.collections import PatchCollection
from matplotlib.path import Path
from matplotlib.patches import PathPatch
from functools import partial

mycmp = 'coolwarm'

df = pd.DataFrame()
df['X'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))
df['Y'] = list(np.random.randint(100, size=100)) + list(np.random.randint(30, size=100))

df['Bin'] = df.apply(lambda row: .1 if row['X'] < 30 and row['Y'] < 30 else .9, axis=1)

#fig, ((ax1, ax2)) = plt.subplots(1, 2, sharex=True, sharey=True)
ax1 = plt.scatter(df['X'], df['Y'])


fig,ax2 = plt.subplots(figsize=(10,10))
hexbin = ax2.hexbin(df['X'], df['Y'], C=df['Bin'], gridsize=15,edgecolors='black',cmap= newcmp , reduce_C_function=np.bincount)
hexbin2 = ax2.hexbin(df['X'], df['Y'], C=df['Bin'], gridsize=15,edgecolors='black',cmap= newcmp , reduce_C_function=np.mean)

def sized_hexbin(ax,hc,hc2):
    offsets = hc.get_offsets()
    orgpath = hc.get_paths()[0]
    verts = orgpath.vertices
    values1 = hc.get_array()
    values2 = hc2.get_array()
    ma = values1.max()
    patches = []

    for offset,val in zip(offsets,values1):
        # Adding condition for minimum size 
        if (val/ma) < 0.2:
            val_t = 0.2
        else:
            val_t = val/ma
        v1 =  verts*val_t + offset
        path = Path(v1, orgpath.codes)
        print(path)
        patch = PathPatch(path)
        patches.append(patch)

    pc = PatchCollection(patches, cmap=  newcmp)  #edgecolors='black'
    pc.set_array(values2)

    ax.add_collection(pc)
    hc.remove()
    hc2.remove()


sized_hexbin(ax2,hexbin,hexbin2)
cb = plt.colorbar(hexbin2, ax=ax2)

plt.xlim((-5, 100))
plt.ylim((-5, 100))

plt.show()