A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.
Python artificial-intelligencedagdata-sciencedata-visualizationdataflowdeveloper-toolsmachine-learningnotebookspipelinepythonreactivesqlweb-app
# /// script
# requires-python = ">=3.9"
# dependencies = [
# "pandas",
# "pymde==0.1.18",
# ]
# ///
import marimo
__generated_with = "0.19.7"
app = marimo.App()
@app.cell
def _(mo):
mo.md("""
# Embedding 🇺🇸 Counties
This notebook accompanies chapter 10 of the monograph
[Minimum-Distortion Embedding](https://web.stanford.edu/~boyd/papers/min_dist_emb.html).
🇺🇸 In this example notebook, we use PyMDE to embed and visualize 3,220
US counties, described by their demographic data (collected between 2013-
2017 by an ACS longitudinal survey).
🌎 We then color each county by the fraction of voters who voted for a
democratic candidate in the 2016 presidential election. Interestingly, the
embedding vaguely resembles a map of the US, clustered by political party
preference, though no geographic or party preference data were
used to compute the embedding!
⚡ We use `mo.ui.altair_chart` to create a reactive and interactive
scatter plot of the embedding: this makes it possible to see where counties
land in the embedding!
""")
return
@app.cell
def _():
import pymde
return (pymde,)
@app.cell
def _(pymde):
dataset = pymde.datasets.counties()
return (dataset,)
@app.cell
def _(mo):
mo.md("""
## The data
The data we embed includes demographic information about each county.
""")
return
@app.cell
def _(dataset, mo):
mo.ui.table(dataset.county_dataframe, page_size=5)
return
@app.cell
def _(mo):
mo.md("""
## The embedding
We now make a neighbor-preserving embedding, to explore the local
relationships in the data.
""")
return
@app.cell
def _(dataset, pymde):
mde = pymde.preserve_neighbors(data=dataset.data, verbose=True)
return (mde,)
@app.cell
def _(mde):
embedding = mde.embed()
return (embedding,)
@app.cell
def _(mo):
mo.md("""
Finally we visualize the embedding, rotating it so that it vaguely
resembles a map of the US. Note that counties that voted Republican tend
to cluster together, as do counties that voted Democratic, even though
our original data had no information about political party preference!
In some real sense, the embedding "discovered" political preference
from demographic data.
""")
return
@app.cell
def _(embedding, pymde):
# Rotate the embedding by some amount of degrees
rotated_embedding = pymde.rotate(embedding, -30.0)
return (rotated_embedding,)
@app.cell
def _(dataset, pd, rotated_embedding):
embedding_df = pd.concat(
[
pd.DataFrame(
{
"x": rotated_embedding[:, 0],
"y": rotated_embedding[:, 1],
"democratic_fraction": dataset.attributes[
"democratic_fraction"
],
}
),
dataset.county_dataframe,
],
axis=1,
)
return (embedding_df,)
@app.cell
def _(mo):
mo.md("""
### Try it! 👆
Select points in the scatter plot below with your cursor: they're
automatically sent back to Python, letting you investigate further!
""")
return
@app.cell
def _(alt, embedding_df, mo):
plot = mo.ui.altair_chart(
alt.Chart(data=embedding_df, width=400, height=400)
.mark_circle(size=10, opacity=1)
.encode(
x="x",
y="y",
color=alt.Color("democratic_fraction").scale(scheme="redblue"),
)
)
plot
return (plot,)
@app.cell
def _(mo, plot):
mo.ui.table(plot.value)
return
@app.cell
def _():
import matplotlib.pyplot as plt
return
@app.cell
def _():
import pandas as pd
return (pd,)
@app.cell
def _():
import altair as alt
return (alt,)
@app.cell
def _():
import marimo as mo
return (mo,)
if __name__ == "__main__":
app.run()