classgraphic

Interactive classification diagnostic plots for scikit-learn.

We classify things for the purpose of doing something to them. Any classification which does not assist manipulation is worse than useless. - Randolph S. Bourne, "Education and Living", The Century Co (April 1917)

Major features:

Plotly based tables for:

• class_imbalance_table
• classification_table
• confusion_matrix_table
• describe (dataframe stats)
• prediction_table
• table

And the following charts:

• class_imbalance
• class_error
• det
• feature_importance
• missing
• precision_recall
• roc
• prediction_histogram
• threshold

For clustering:

• Delauney triangulations
• Voronoi tessalations

Try it

By trying it on binder, you'll see all the details and interactivity. The quickstart below has static images, but if you run these commands in a jupyter notebook, ipython or IDE you will be able to interact with them.

Quickstart

from classgraphic.essential import *

# loading the data
df = px.data.iris()

# let's see what kind of data we have
describe(df, transpose=True).show()

# any missing?
missing(df)

# features
X = df.drop(columns=["species", "species_id"])

#target
y = df["species"]

# Let's check our classes we will be training on and predicting
class_imbalance_table(y, condition="all")

# train / test split
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.5, random_state=random_state
)

# we want to see total count for each, default for bars is to be stacked, so that works
# we could also pass to class_imbalance barmode="overlay" if we prefer
class_imbalance(y_train, y_test, condition="train,test")

# model
model = LogisticRegression(max_iter=max_iter, random_state=random_state)
model.fit(X_train, y_train)

# predictions
y_score = model.predict_proba(X_test)
y_pred = model.predict(X_test)

confusion_matrix_table(model, y_test, y_pred).show()
classification_table(model, y_test, y_pred)

feature_importance(model, y, transpose=True)

This concludes the quickstart. There are many more visualizations and tables to explore.

See the notebooks and docs folders on github and the documentation web site for more information.

Requirements

• Python 3.8 or later
• numpy
• pandas
• plotly>=5.0
• scikit-learn
• nbformat

Install

If you use conda, create an environment named classgraphic, then activate it:

• in Linux: source activate pilot

• In Windows: conda activate pilot

If you use another environment management create and activate your environment using the normal steps.

Then execute:

python setup.py install

or for installing in development mode:

python -m pip install -e . --no-build-isolation

or alternatively

python setup.py develop

To install from github instead:

pip install git+https://github.com/dionresearch/classgraphic

See also

• stemgraphic python package for visualization of data and text
• Hotelling one and two sample Hotelling T2 tests, T2 and f statistics and univariate and multivariate control charts and anomaly detection

History

0.3.1 (2023-09-20)

• bugfix for describe with pandas 2.x

0.3.0 (2023-05-01)

• added 2D clustering visualization
• defaults to Voronoi tessalation
• optional Delauney triangulation

0.2.1 (2022-09-20)

• fixed image not showing on pypi
• fixed feature importance error
• warning = False didn't prevent warning to show

0.2.1 (2022-09-19)

• added binary classification notebook example
• fixed issue with non dataframe binary classification

0.2.0 (2022-09-18)

The previous version was a first step to doing a public release. This release added:

• documented
• updated the code to be in line with plotly 5.x

It was released to github and pypy.

0.1.0 (2019-10-27)

• First private release

Origins

Inspired by Dion Research LLC Internal EDA/anomaly and end to end data science platform. A dozen charts and tables were initially designed to provide better diagnostic reporting. Some can also be used for exploratory or explanatory purposes.

See: https://blog.dionresearch.com/2019/10/visualizations-explanatory-exploratory.html