caching.md

Improve the performance with Caching

One of the key concepts in Streamlit is caching.

In Streamlit

• your script is run once when a user visits the page.

• your script is rerun top to bottom on user interactions.

Thus with Streamlit you must use caching to make the user experience nice and fast.

In Panel

• your script is run once when a user visits the page.

• only specific, bound functions are rerun on user interactions.

Thus with Panel you may use caching to make the user experience nice and fast.

In Panel you use pn.cache to speed up your apps. Check out the Cache How-To Guides for more details.

---

Migration Steps

To migrate

• replace st.cache_data and st.cache_resource with pn.cache on long running

  • functions that are run when your page loads

  • bound functions

Example

Cache Example

Streamlit Cache Example

from time import sleep

import numpy as np
import streamlit as st
from matplotlib.figure import Figure

@st.cache_data
def get_data():
    print("get_data func")
    sleep(1.0)
    return np.random.normal(1, 1, size=100)

@st.cache_data(hash_funcs={Figure: lambda _: None})
def plot(data, bins):
    print("plot func", bins)
    sleep(2)
    fig = Figure(figsize=(8,4))
    ax = fig.subplots()
    ax.hist(data, bins=bins)
    return fig

data = get_data()
bins = st.slider(value=20, min_value=10, max_value=30, step=1, label="Bins")
st.pyplot(plot(data, bins))

I've added sleep statements to make the functions more expensive.

Panel Cache Example

from time import sleep

import numpy as np
import panel as pn
from matplotlib.figure import Figure

@pn.cache
def get_data():
    print("get_data func")
    sleep(1.0)
    return np.random.normal(1, 1, size=100)

@pn.cache
def plot(data, bins):
    print("plot func", bins)
    sleep(2)
    fig = Figure(figsize=(8,4))
    ax = fig.subplots()
    ax.hist(data, bins=bins)
    return fig

pn.extension(sizing_mode="stretch_width", template="bootstrap")

data = get_data()
bins = pn.widgets.IntSlider(value=20, start=10, end=30, step=1)
bplot = pn.bind(plot, data, bins)
pn.Column(bins, pn.panel(bplot, loading_indicator=True)).servable()

You can also use pn.cache as an function. I.e. as

plot = pn.cache(plot)

Using pn.cache as a function can help you keep your business logic (data and plot function) and your caching logic (when and how to apply caching) separate. This can help you reusable and maintainable code.