Pydeck hexlayer slow with large number of points - streamlit

Question

Let's consider this simple example of HexagonLayer with a large number of points:
import numpy as np
import pandas as pd
import pydeck as pdk
import streamlit as st
lat0=40.7
lon0=-74.12
n_points = 100000
lat = np.random.normal(loc=lat0, scale=0.02, size=n_points)
lon = np.random.normal(loc=lon0, scale=0.02, size=n_points)
data = pd.DataFrame({'lat': lat, 'lon': lon})
st.pydeck_chart(pdk.Deck(
map_provider="mapbox",
initial_view_state=pdk.ViewState(
latitude=lat0,
longitude=lon0,
zoom=10,
),
layers=[
pdk.Layer(
'HexagonLayer',
data=data,
get_position='[lon, lat]',
radius=1000,
coverage=0.6,
),
],
))
Here's the result:
AFAIU, HexagonLayer is a histogram of the data. Here, I have 100k points in the dataframe, but only o(100) non empty bins.
When moving the map or zooming on the map, things are painfully slow. It looks as if the histogramming is reran at every step (though with the same result, as opposed to other layers types like ScreenGridLayer or HeatmapLayer). Is there a way to cache the result of the hexagonal binning to speed things up? Or am I doing something wrong?
As a test, if I reduce the number of points, the map becomes way more usable (even if I increase the number of non empty bins a lot).