Practical overview
When it fits
Use Tensorwatch during development when you want to inspect training activity in Jupyter or build custom views for model and data analysis.
Who it is for
- Python data-science practitioners
- Deep-learning and reinforcement-learning developers
- ML developers who want live training visualizations or custom analysis dashboards
Documented uses
- View changing training values in a Jupyter notebook as they are written.
- Query a live training process and visualize the resulting stream.
- Build custom visualizations, interfaces and dashboards for models and data.
- Connect streams such as files, sockets and visualizers into data-flow graphs.
Requirements & limits
The repository is public and lists an MIT license. The README gives `pip install tensorwatch` as the installation method.
- The documented workflow centers on Python and Jupyter Notebook.
- The README lists Python 3.x and testing with PyTorch 0.4–1.x; it says most features should also work with TensorFlow eager tensors.
- Graphviz is used for network diagrams and may require separate installation depending on platform.
- The project warns that Lazy Logging can execute arbitrary Python expressions and says Tensorwatch is not designed for production, multi-tenant or adversarial environments.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Create a live metric graph in Jupyter
Log values to a Tensorwatch stream and view them as a live line graph in a generated notebook.
Before you start
- Python 3.x
- Tensorwatch installed with pip
- Jupyter Notebook
Steps
Install Tensorwatch with the command below.
pip install tensorwatchIn Python, create a
Watcherwith a filename, create a named stream, and write values to it; the documented example writes a tuple every second.Call
make_notebook()on the watcher to create the notebook file.Open the generated notebook in Jupyter and run all cells to view the line graph as values are written.
Expected outcome
A Jupyter notebook displays a live graph of values written to the Tensorwatch stream.
Guide limitations
- The documented example uses a generated test stream rather than a model-specific training integration.
- Graphviz may need separate installation on some platforms for network diagrams.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 10 October 2026
- Original website title
- microsoft/tensorwatch · Official homepage title · retained in its source language