Can I smooth out noisy loss curves in the generated plots?

Yes, the skill provides several smoothing options, including Exponential Moving Average (EMA) weight adjustment and simple rolling average windows.

Can I compare multiple experiment runs in a single plot?

Yes, the generate_plots script supports comma-separated run IDs, allowing you to overlay metrics from multiple runs on the same chart for easy comparison.

Where are the plots and metadata stored?

By default, the skill saves images and a metadata.json file in a structured directory format: wandb_plots/[entity]_[project]/[run_name]_[run_id]/.

What happens if a run doesn't have an existing plot image?

If download_plots finds no existing images, you can use the generate_plots command to create new visualizations directly from the recorded metric history.

How do I authenticate with Weights & Biases for this skill?

You should set the WANDB_API_KEY environment variable (recommended) or run the 'wandb login' command in your terminal before using the skill.

Weights & Biases Plotting

Name: Weights & Biases Plotting
Author: RubenBranco

byRubenBranco

数据科学与机器学习

Downloads and generates high-quality metric visualizations from Weights & Biases experiment runs directly within Claude.

关于

The wandb-plot skill bridges the gap between experiment tracking and analysis by allowing users to interact with Weights & Biases data seamlessly. It provides a comprehensive suite of tools to browse projects, list runs, and inspect specific metrics, as well as the ability to download existing run artifacts or generate fresh line charts for performance analysis. Supporting advanced features like EMA smoothing, multi-run comparison, and viewport scaling, this skill is designed for machine learning researchers and engineers who need to visualize training progress and compare model iterations without context-switching from their workspace.

主要功能

Inspect detailed metric statistics including min, max, and mean
Generate custom line charts for metrics like loss and accuracy
0 GitHub stars
Browse and list W&B projects and runs
Apply EMA smoothing and viewport scaling to generated plots
Download existing plot images and media artifacts from runs

使用场景

Comparing training performance across multiple model architectures or hyperparameters
Exporting high-quality metric charts for research documentation and status reports
Visualizing training stability and system metrics for machine learning experiments

关于

主要功能

Inspect detailed metric statistics including min, max, and mean
Generate custom line charts for metrics like loss and accuracy
0 GitHub stars
Browse and list W&B projects and runs
Apply EMA smoothing and viewport scaling to generated plots
Download existing plot images and media artifacts from runs

使用场景

Comparing training performance across multiple model architectures or hyperparameters
Exporting high-quality metric charts for research documentation and status reports
Visualizing training stability and system metrics for machine learning experiments