cluestar 0.1.0

Gain a clue by clustering!

cluestar

Gain a clue by clustering!

This library contains visualisation tools that might help you get started with classification tasks. The idea is that if you can inspect clusters easily, you might gain a clue on what good labels for your dataset might be!

It generates charts that looks like this:

Install

python -m pip install "cluestar @ git+https://github.com/koaning/cluestar.git"

Interactive Demo

You can see an interactive demo of the generated widgets here.

You can also toy around with the demo notebook found here.

Usage

The first step is to encode textdata in two dimensions, like below.

from sklearn.pipeline import make_pipeline
from sklearn.decomposition import TruncatedSVD
from sklearn.feature_extraction.text import TfidfVectorizer

pipe = make_pipeline(TfidfVectorizer(), TruncatedSVD(n_components=2))

X = pipe.fit_transform(texts)

From here you can make an interactive chart via;

from cluestar import plot_text

plot_text(X, texts)

The best results are likely found when you use umap together with something like universal sentence encoder.

You might also improve the understandability by highlighting points that have a certain word in it.

plot_text(X, texts, color_words=["plastic", "voucher", "deliver"])

You can also use a numeric array, one that contains proba-values for prediction, to influence the color.

# First, get an array of pvals from some model
p_vals = some_model.predict(texts)[:, 0]
# Use these to assign pretty colors.
plot_text(X, texts, color_array=p_vals)