tensorwatch
Debugging, monitoring and visualization for Python Machine Learning and Data Science
What is tensorwatch?
Debugging, monitoring and visualization for Python Machine Learning and Data Science
Overview of tensorwatch
What it is
TensorWatch is a debugging and visualization tool designed for data science, deep learning and reinforcement learning from Microsoft Research. It works in Jupyter Notebook to show real-time visualizations of machine learning training and perform analysis tasks for models and data. It is designed to be flexible and extensible, allowing custom visualizations, UIs, and dashboards. It includes a Lazy Logging Mode that executes arbitrary queries against live ML training and returns a stream as a result.
Key features
- Real-time visualization in Jupyter Notebook (line graphs, histograms, pie charts, scatter charts, etc.)
- Lazy Logging Mode: execute arbitrary Python expressions against live training process and view results with chosen visualizer - Data flow graph architecture: streams can listen to any other streams, enabling arbitrary data flow graphs - Supports Python 3.x, PyTorch 0.4-1.x, and TensorFlow eager tensors - Security mitigations: HMAC-SHA256 authentication, RestrictedUnpickler with allowlist, localhost binding by default - Can generate Jupyter Notebook automatically from script
Use cases
Debugging and monitoring machine learning training, real-time visualization in Jupyter Notebook, building custom dashboards, interactive debugging via lazy logging, analyzing model and data.
Getting started
pip install tensorwatch Example code from the README: import tensorwatch as tw import time w = tw.Watcher(filename='test.log') s = w.create_stream(name='metric1') w.make_notebook() for i in range(1000): s.write((i, i*i)) time.sleep(1) When run, a Jupyter Notebook file test.ipynb is created. Open it with jupyter notebook and run all cells to see real-time line graph.
License and pricing
MIT license, open source.
Based on the project's public README and repository metadata.