Visdom

Visdom is an open-source visualization tool developed by Facebook AI Research (FAIR) for real-time visualization of data during machine learning experiments. It provides a Python client library and a lightweight web server that serves a browser-based interface, allowing researchers to monitor training progress, inspect model outputs, and explore data interactively.

Visdom operates on a client-server model. The Python library creates visualization windows and streams data to the server, which renders them in a web browser. It supports a range of visualizations, including 2D line plots, scatter plots, bar charts, heatmaps, images, audio, and plain text. Windows are identified by an integer or string name and can be updated incrementally, enabling live plotting and dynamic dashboards during experiments.

Usage typically involves installing the package with pip, starting the server with python -m visdom.server, and connecting to the UI at a local URL such as http://localhost:8097. In the Python environment, a Visdom instance is created and used to create and update visualizations, for example by plotting training metrics as they are produced.

Visdom was released as an open-source project by FAIR to provide an accessible alternative for real-time visualization in ML workflows, and it has been widely used in PyTorch and other research settings. While still usable, it faces competition from other tools and dashboards, and some users prefer more feature-rich or actively maintained options today.