visdom

Visdom is a browser-based visualization tool for data scientists and researchers who want to monitor experiments as they run. You run a small server locally or on a remote machine, and it serves a web dashboard where you can see charts, images, and text that your code sends to it in real time. The typical use case is watching a machine learning training run: as your model trains, you push metrics like loss and accuracy to Visdom, and the dashboard updates automatically so you can see progress without checking log files.

The workspace is organized around "windows," which are draggable and resizable panels that each contain one plot, image, or text block. You can arrange these panels however you like, and the layout is saved across sessions. A separate concept called "environments" lets you group windows into distinct workspaces, one per experiment or project, and switch between them in the UI. Multiple environments can also be overlaid for side-by-side comparison of results from different runs.

Charts can be built using many common types: line charts, scatter plots, bar charts, heatmaps, histograms, and more. Images can be sent as numpy arrays or tensors and displayed in a grid. The server can be reached from any browser, so if you are running a long training job on a remote machine, a teammate can view the same dashboard by visiting the server URL.

Callbacks let Python code react to things that happen in the UI, such as a button click or a key press inside an image panel. Plot parameters can also be edited live in the interface without rerunning code. The tool is maintained by the FOSSASIA open source community and supports Python through a pip-installable client library.

The full README is longer than what was shown.