Custom Charts

2 minute read

The W&B Python SDK includes functions for creating specialized charts from your data. These functions generate interactive visualizations that integrate with the W&B UI.

Overview

Custom Charts in W&B (wandb.plot) are visualization functions that transform data into interactive charts. These functions handle common ML visualization requirements such as confusion matrices, ROC curves, and distribution plots. Custom charts can also be created using Vega-Lite specifications.

Available Chart Functions

Function	Description
`confusion_matrix()`	Generate confusion matrices for classification performance visualization.
`roc_curve()`	Create Receiver Operating Characteristic curves for binary and multi-class classifiers.
`pr_curve()`	Build Precision-Recall curves for classifier evaluation.
`line()`	Construct line charts from tabular data.
`scatter()`	Create scatter plots for variable relationships.
`bar()`	Generate bar charts for categorical data.
`histogram()`	Build histograms for data distribution analysis.
`line_series()`	Plot multiple line series on a single chart.
`plot_table()`	Create custom charts using Vega-Lite specifications.

Common Use Cases

Model Evaluation

Classification: confusion_matrix(), roc_curve(), and pr_curve() for classifier evaluation
Regression: scatter() for prediction vs. actual plots and histogram() for residual analysis

Training Monitoring

Learning Curves: line() or line_series() for tracking metrics over epochs
Hyperparameter Comparison: bar() charts for comparing configurations

Data Analysis

Distribution Analysis: histogram() for feature distributions
Correlation Analysis: scatter() plots for variable relationships

Custom Visualizations

Vega-Lite Charts: plot_table() for domain-specific visualizations
Multi-Chart Dashboards: Combination of multiple chart types in a single run

Example: Scatter plot

import wandb
import numpy as np
from sklearn.metrics import roc_curve as sklearn_roc_curve

# Initialize a run
wandb.init(project="custom-charts-demo")

# Example 1: Log a confusion matrix
y_true = [0, 1, 2, 0, 1, 2]
y_pred = [0, 2, 2, 0, 1, 1]
class_names = ["class_0", "class_1", "class_2"]

wandb.log({
    "conf_mat": wandb.plot.confusion_matrix(
        y_true=y_true, 
        preds=y_pred,
        class_names=class_names
    )
})

# Example 2: Create an ROC curve for binary classification
y_true_binary = np.array([0, 0, 1, 1, 0, 1])
y_scores = np.array([0.1, 0.4, 0.35, 0.8, 0.2, 0.9])

wandb.log({
    "roc": wandb.plot.roc_curve(
        y_true=y_true_binary, 
        y_probas=y_scores,
        labels=["negative", "positive"]
    )
})

# Example 3: Create a custom line chart from a table
table = wandb.Table(columns=["epoch", "accuracy", "loss"])
for epoch in range(10):
    table.add_data(epoch, 0.8 + 0.02 * epoch, 1.0 - 0.1 * epoch)

wandb.log({
    "training_progress": wandb.plot.line(
        table, x="epoch", y="accuracy",
        title="Model Accuracy Over Time"
    )
})

# Example 4: Build a scatter plot for feature analysis
data_table = wandb.Table(columns=["feature_1", "feature_2", "label"])
for _ in range(100):
    data_table.add_data(
        np.random.randn(), 
        np.random.randn(), 
        np.random.choice(["A", "B"])
    )

wandb.log({
    "feature_scatter": wandb.plot.scatter(
        data_table, x="feature_1", y="feature_2",
        title="Feature Distribution"
    )
})

wandb.finish()