Penguin Kmeans

This app shows k-means clustering of the Palmer Penguins dataset with scikit-learn. The number of clusters and the plotted variables are configurable, and brushing on the Altair chart filters the linked table.

import altair as alt
import pandas as pd
import panel as pn

from sklearn.cluster import KMeans

pn.extension('tabulator', 'vega')

Load data

penguins = pn.cache(pd.read_csv)('https://datasets.holoviz.org/penguins/v1/penguins.csv').dropna()
cols = list(penguins.columns)[2:6]

Define application

Only the clustering is cached. The chart is cheap to build and Altair charts are mutable, so sharing one between sessions is not safe.

brush = alt.selection_interval(name='brush')

@pn.cache
def get_clusters(n_clusters):
    kmeans = KMeans(n_clusters=n_clusters, n_init='auto')
    est = kmeans.fit(penguins[cols].values)
    df = penguins.copy()
    df['labels'] = est.labels_.astype('str')
    return df

def get_chart(x, y, df):
    centers = df.groupby('labels')[[x] if x == y else [x, y]].mean()
    return (
        alt.Chart(df)
            .mark_point(size=100)
            .encode(
                x=alt.X(x, scale=alt.Scale(zero=False)),
                y=alt.Y(y, scale=alt.Scale(zero=False)),
                shape='labels',
                color='species'
            ).add_params(brush) +
        alt.Chart(centers)
            .mark_point(size=250, shape='cross', color='black')
            .encode(x=x+':Q', y=y+':Q')
    ).properties(width='container', height='container')

def vega_filter(filters, df):
    filtered = df
    for field, drange in (filters or {}).items():
        filtered = filtered[filtered[field].between(*drange)]
    return filtered

intro = pn.ui.Markdown("""
This app provides an example of **building a simple dashboard using
Panel**.\n\nIt demonstrates how to take the output of **k-means
clustering on the Penguins dataset** using scikit-learn,
parameterizing the number of clusters and the variables to
plot.\n\nThe plot and the table are linked, i.e. selecting on the plot
will filter the data in the table.\n\n The **`x` marks the center** of
the cluster.
""", sizing_mode='stretch_width')

x = pn.ui.Select(label='x', options=cols, value='bill_depth_mm')
y = pn.ui.Select(label='y', options=cols, value='bill_length_mm')
n_clusters = pn.ui.IntSlider(label='n_clusters', start=1, end=5, value=3)

clusters = pn.bind(get_clusters, n_clusters)

chart = pn.ui.Vega(
    pn.bind(get_chart, x, y, clusters), min_height=400, max_height=800, sizing_mode='stretch_width'
)

table = pn.ui.Tabulator(
    clusters,
    pagination='remote', page_size=10, height=600,
    sizing_mode='stretch_width'
)

table.add_filter(pn.bind(vega_filter, chart.selection.param.brush))

Layout app

page = pn.ui.Page(
    main=[intro, chart, table],
    sidebar=[x, y, n_clusters],
    title="KMeans Clustering"
)

page

Served App

if pn.state.served:
    page.servable()