Interactive hierarchical clustering

Using Streamlit
,
1

Hi,

I want to display an interactive dendrogram (a tree representing the result of the an hierarchical clustering), such that when i put the cursor over a node – either a leaf of an inner node – I can see some labels corresponding to leaves within the subtree rooted in the node.

What I currently have:

import time
import numpy as np
import streamlit as st
import matplotlib
from matplotlib import pyplot as plt
matplotlib.use(‘TkAgg’)

progress_bar = st.sidebar.progress(0)
status_text = st.sidebar.empty()

import time
import numpy as np
import streamlit as st
import matplotlib
from matplotlib import pyplot as plt
matplotlib.use(‘TkAgg’)
random.seed(0)
np.random.seed(0)
progress_bar = st.sidebar.progress(0)
status_text = st.sidebar.empty()

from scipy.cluster.hierarchy import linkage, dendrogram
from matplotlib import pyplot as plt
data = np.random.rand(1000,30) - 0.5
labels = [“a”, “b”, “c”, “d”, “e”, “f”," ,“g”, “h”, “i”, “j”]
Z = linkage(data, optimal_ordering = False, method = ‘ward’)
dn = dendrogram(Z, truncate_mode = ‘lastp’)
st.pyplot()
print(“Here”)
progress_bar.empty()

Streamlit widgets automatically run the script from top to bottom. Since

this button is not connected to any other logic, it just causes a plain

rerun.

st.button(“Re-run”)

I want to be able to put the cursor over the leaf corresponding to the label "a", and see that label; or put the cursor over the inner node which is the ancestor of the leaves "a", "e", "h" and "j", and see those labels (see image).

illustration
illustration782×580 15.9 KB

Thanks!

2

Hey,

I’m not sure it’s easy to do interactive charting with Matplotlib, but Streamlit can also display plotly which has an interactive tree-plot. It’s the only interactive tree I could find within the supported interactive plotting options for Streamlit.

That would need figuring out how to transpose the dendogram leaves output into plotly input though, maybe that could be a good strarting point.

3

Hi @Shauli_Ravfogel,
definitely you can try out other plotting libraries that are offered in Streamlit. I think altair and bokeh also support interactive chart, but you will have to transpose your model as suggested by @andfanilo.

Best,
Matteo