tbparse

What it is and what it does

tbparse is a parser that reads TensorBoard event log files and converts them into pandas DataFrames, making it easy to programmatically access and analyze training metrics and other logged data. It supports all major event types (scalars, tensors, histograms, images, audio, hparams, text) and works with logs generated by PyTorch, Keras/TensorFlow, and TensorboardX. The package stores parsed data in DataFrames to enable filtering, aggregation, and custom analysis workflows outside the TensorBoard dashboard.

The package is designed for researchers and engineers who need to extract and manipulate TensorBoard logs programmatically—for example, to compare experiments across runs, perform statistical analysis, or generate custom plots. It requires only pandas and tensorboard as runtime dependencies, with TensorFlow optionally needed for image and audio parsing. The codebase includes full type hints and high test coverage.

Use it for:

• Extract training scalars (loss, accuracy) from multiple experiment runs and compare them in a single DataFrame.
• Programmatically filter and aggregate hyperparameter search results logged via TensorBoard.
• Parse tensorboard event logs into DataFrames for statistical analysis or custom visualization outside the dashboard.
• Batch process tensorboard logs from a hierarchical directory structure of experiments.
• Build automated pipelines to extract metrics from PyTorch or TensorFlow training runs for downstream reporting.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Parses TensorBoard event log files and loads them as pandas DataFrames, supporting scalars, tensors, histograms, images, audio, hparams, and text from PyTorch, Keras/TensorFlow, and TensorboardX.

Yes, if you need to programmatically read and analyze TensorBoard logs. The package is stable, has low install friction, permissive licensing, and no known vulnerabilities. However, note that maintenance is dormant (last release 728 days ago), so compatibility with very recent TensorBoard versions is not guaranteed. It is well-suited for one-off analysis or established workflows but may require monitoring if you adopt it for new projects with cutting-edge tools.

Install

tbparse on PyPI

pip install tbparse

uv add tbparse

poetry add tbparse

Installing tbparse

Before you install

Low install friction with a pure-Python wheel. Maintenance is dormant—last release was 728 days ago—but the repository remains active and the package supports current Python versions (3.8–3.12).

License in practice

Distributed under Apache License 2.0 (permissive). Runtime dependencies pandas (BSD-3-Clause) and tensorboard (Apache 2.0) are also permissive, so there are no licensing restrictions on use or redistribution.

Quickstart

pip install tbparse

from tbparse import SummaryReader
reader = SummaryReader("<path_to_event_file_or_directory>")
df = reader.scalars
print(df)

Requires Python >= 3.8. Parsing images and audio requires TensorFlow as an optional dependency.

Verify before relying

• Whether dormant status (728 days since last release) will affect long-term compatibility with newer TensorBoard versions.
• Performance characteristics when parsing very large event files or deeply nested directory structures.

Package facts

Evidence: tbparse-0.0.9-py3-none-any.whl

Keywords: package, parser, plot, python, pytorch, reader, tensorboard, tensorboardx, tensorflow

Tags

More Scientific/Engineering packages