Introduction

When creating a web application with graphs based on Python, I have used the combination of the graph library Plotly + web application framework Dash by Plotly. However, this time I met a large rookie Streamlit with great potential to threaten the position of Dash, so I will drop a memo written as a test.

What is `Streamlit`?

Streamlit — The fastest way to build custom ML tools

Streamlit is an open-source app framework for Machine Learning and Data Science teams. Create beautiful data apps in hours, not weeks. All in pure Python. All for free.

An application framework that allows you to create a web application with a single Python script without writing HTML / CSS / JavaScript. The front end is made with React + Bootstrap, and it can be said that it is almost the same type of framework as Dash, including the point that the Python script alone is OK.

Compared to Dash, Streamlit does not have the flexibility to freely decide the style and page layout. There is no API to put custom CSS and style, and the page layout is firmly decided by the side menu + body template. In other words, since the framework side defines the specifications including the design, you can focus more than Dash on writing scripts for the main processing of data processing and analysis without the cost of designing.

Also, Dash is a web application framework dedicated to Plotly, but Streamlit supports the following graph libraries.

Matplotlib --Seaborn is also a wrapper for Maptplotlib, so it's probably okay
Altair
Graphviz
Plotly ――Of course, Plotly Express is okay too!
Bokeh

Plotly is not so much, you can create a web application even in any case.

Test writing memo

Installation

$ pip install streamlit

Demo page launch

$ streamlit hello

The demo page will be launched at http: // localhost: 8501 /.

Write a script

Write a sample script qiita.py that plots the data for Plotly Express. It's pretty simple.

`qiita.py`


import streamlit as st
import plotly.express as px
import plotly.io as pio

# data
data = px.data.iris()

# sidemenu
st.sidebar.markdown(
    "# Qiita sample"
)
template = st.sidebar.selectbox(
    "Template", list(pio.templates.keys())
)

# body
st.write(
    px.scatter(data, x="sepal_width", y="sepal_length", template=template)
)

On the directory where qiita.py is located

$ streamlit run qiita.py

Will launch the application at http: // localhost: 8501 /.

If you write in Dash, you can write HTML components in a hurry, and I think that the amount of code will be doubled.

Tips

I will update it from time to time.

If you really want to modify the style

Create a markdown component with st.markdown and write the<style>tag directly while giving true to the ʻunsafe_allow_html` parameter.

st.markdown(
    "<style>h1{color: red}</style>",
    unsafe_allow_html=True
)

It's a black magic code because Streamlit doesn't have a styling mouth, but even if you check the forum page of the main body, it seems that there is only this much,

Remote resource `streamlit run`

For example, if you want to execute the source code uploaded to GitHub quickly, Streamlit can be executed as follows without a local clone.

$ streamlit run https://raw.githubusercontent.com/prs-watch/streamlit-sample/master/sample.py

The sample repository illustrated above is here.

Summary

Since the cost of designing can be cut down, you can create a web application quickly with a small amount of code. The design provided is simple and I like it. Dash has the good points of Dash, and Streamlit itself has not been released in the major version, but it is quite useful if you want to create an application easily.

The recommendation level is high, so if you want to make a dashboard, for example, why not write a test once?

[Python] I wrote a test of "Streamlit" that makes it easy to create visualization applications.