Scikit-Plot

Warnings

• breaking The Factory API, a previous way to instantiate plots, was deprecated in v0.3.0 and announced for removal in v0.4.0. Direct function calls from specific modules should be used instead.
  Fix: Migrate to the Functions API, importing plotting functions directly from `skplt.metrics`, `skplt.estimators`, `skplt.clusterer`, or `skplt.decomposition`.
• deprecated Functions `plot_precision_recall_curve` and `plot_roc_curve` were deprecated in v0.3.5. They have been replaced by `plot_precision_recall` and `plot_roc`, respectively, which offer more control over plotted curves.
  Fix: Replace `skplt.metrics.plot_precision_recall_curve` with `skplt.metrics.plot_precision_recall` and `skplt.metrics.plot_roc_curve` with `skplt.metrics.plot_roc`.
• deprecated The `scikitplot.plotters` module was deprecated in v0.3.0, and its functions were distributed to more specialized modules (e.g., `skplt.metrics`, `skplt.estimators`).
  Fix: Update import statements to directly reference the new module paths, e.g., `from scikitplot.metrics import plot_confusion_matrix` (or `import scikitplot as skplt; skplt.metrics.plot_confusion_matrix`).
• gotcha The `spectral` colormap used internally was deprecated and changed to `nipy_spectral` in v0.3.2 to avoid Matplotlib warnings/errors.
  Fix: Upgrade to scikit-plot v0.3.2 or later, or manually specify a different colormap where applicable if using older versions.
• gotcha As of its latest release (v0.3.7, August 2018), `scikit-plot` was built for Python versions and `scikit-learn` versions prevalent at that time (e.g., Python 3.5-3.7). Using it with very recent Python or `scikit-learn` versions (e.g., Python 3.9+ or scikit-learn 1.0+) may lead to compatibility issues or require specific dependency pinning.
  Fix: Consider running `scikit-plot` within an environment that matches its original dependencies, typically Python 3.5-3.8 and a compatible `scikit-learn` version (e.g., <1.0). Alternatively, evaluate `scikit-plots` (note the plural), a potentially distinct and more recently maintained library with similar goals, if targeting newer Python environments.

Install

• pip install scikit-plot Install stable version

Imports

• skplt.metrics.plot_confusion_matrix

  import scikitplot as skplt
  skplt.metrics.plot_confusion_matrix(y_true, y_pred)

  The `scikitplot.plotters` module was deprecated in v0.3.0; functions were moved to specific submodules like `metrics`, `estimators`, etc.
• skplt.metrics.plot_roc

  import scikitplot as skplt
  skplt.metrics.plot_roc(y_true, y_probas)

  `plot_roc_curve` was deprecated in v0.3.5 in favor of `plot_roc`.
• skplt.metrics.plot_precision_recall

  import scikitplot as skplt
  skplt.metrics.plot_precision_recall(y_true, y_probas)

  `plot_precision_recall_curve` was deprecated in v0.3.5 in favor of `plot_precision_recall`.

Quickstart

This quickstart demonstrates how to train a RandomForestClassifier on the digits dataset and then use scikit-plot to visualize its normalized confusion matrix with a single function call. It shows the typical import pattern and the use of a common metric plotting function.

import matplotlib.pyplot as plt
import scikitplot as skplt
from sklearn.datasets import load_digits
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split, cross_val_predict

# Load dataset
X, y = load_digits(return_X_y=True)

# Split data (though cross_val_predict handles internal splits)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

# Train a classifier
clf = RandomForestClassifier(n_estimators=100, random_state=42)
clf.fit(X_train, y_train)

# Get predictions
y_pred = clf.predict(X_test)

# Plot confusion matrix
skplt.metrics.plot_confusion_matrix(y_test, y_pred, normalize=True)
plt.title('Normalized Confusion Matrix')
plt.show()