- TensorWatch provides many features for improving debugging capabilities in both pre- and post-training phases of model development.
- It also supports several standard visualization types, including bar charts, histograms, pie charts as well as 3D variations.
The rise of deep learning is accompanied by ever-increasing model complexity, larger datasets and longer training times for models. But help is at hand to overcome these challenges.
Microsoft Research team brings to us an advanced open source tool for machine learning training that could help researchers make more informed decisions, as well as dramatically reduce the cost of logging.
With TensorWatch—a debugging and visualization tool for machine learning—researchers and engineers can customize the user interface to accommodate a variety of scenarios, Microsoft said.
A research team led by Shital Shah recently announced the open source release of TensorWatch in a blog post.
“We like to think of TensorWatch as the Swiss Army knife of debugging tools with many advanced capabilities researchers and engineers will find helpful in their work,” the team wrote.
How it works?
TensorWatch provides the interactive debugging of real-time training processes using either the composable UI in Jupyter Notebooks or the live shareable dashboards in Jupyter Lab.
In addition, since TensorWatch is a Python library, researchers can also build their own custom UIs or use TensorWatch in the vast Python data science ecosystem. TensorWatch also supports several standard visualization types, including bar charts, histograms, pie charts, as well as 3D variations.
With TensorWatch, the team also introduces lazy logging mode. “This mode doesn’t require explicit logging of all the information beforehand. Instead, you can have TensorWatch observe the variables. Since observing is basically free, you can track as many variables as you like, including large models or entire batches during the training,” they said.
The power of open source
TensorWatch provides many features that can improve debugging capabilities in all phases of model development—pre-training, in-training, and post-training.
Microsoft uses several open source libraries to enable many of these features, which include model graph visualization, data exploration through dimensionality reduction, model statistics and several prediction explainers for convolution networks.