A debugging and visualization tool for data science, deep learning, and reinforcement learning in Jupyter Notebook.
TensorWatch is a debugging and visualization tool for machine learning and data science, developed by Microsoft Research. It enables real-time monitoring of training metrics, interactive querying of live processes, and comprehensive model analysis—all within Jupyter Notebook. It solves the problem of opaque training loops by providing immediate visual feedback and deep insights into model behavior.
Data scientists, machine learning engineers, and researchers who use Python frameworks like PyTorch or TensorFlow and develop models in Jupyter Notebook environments.
Developers choose TensorWatch for its unique lazy logging capability, which allows querying live training without pre-instrumentation, and its extensible stream-based architecture that supports custom visualizations and dashboards beyond standard logging tools.
Debugging, monitoring and visualization for Python Machine Learning and Data Science
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Enables live plotting of training metrics directly within Jupyter Notebook, providing immediate feedback during model development, as shown in the quick start example with real-time line graphs.
Allows dynamic querying of live training processes via Python expressions to generate data streams without pre-logging, a unique feature highlighted in the lazy logging mode tutorial.
Combines tools for visualizing model graphs with tensor shapes, layer statistics (FLOPs, parameters), dataset exploration with t-SNE, and prediction explainers like LIME in one package.
Built on a hackable, stream-based system that supports custom visualizations and dashboards, emphasizing extensibility as per the philosophy section.
Admits to using eval() on network-supplied expressions and pickle deserialization, posing significant threats in untrusted environments, as detailed extensively in the security notice.
Primarily designed for Jupyter Notebook, limiting use in headless servers or alternative interfaces without major modifications, as visualizations are notebook-bound.
Labeled as under heavy development in the README, which may lead to breaking changes, bugs, or incomplete features for users seeking stability.