Skip to main content

A tiny CSV export micro-library for Weights and Biases.

Project description

W&B Summarizer

A humble unofficial microlibrary / command line tool for summarizing data within Weights and Biases across runs including a tiny tool to download ("export") wandb data as a CSV file. This is not endorsed or affiliated with Weights and Biases.


Installation

This package is available via pip and requires Python 3. Simply run the following (optionally within a virtual environment).

$ pip install wandb_summarizer

Usage

This package is usable via the command line or from another Python script. The command line tool takes the following form:

$ wandb-summarizer-to-csv [username/project name] [output loc] [optional verbose flag] [optional query parameters]

Most uses of the microlibrary from another Python script will want to use wandb_summarizer.download.get_results which takes the following parameters:

  • project_name (str): The name of the project to be downloaded. Should be of form username/project.
  • query_param (str): The mongo-like selector to be used in filtering runs. If not provided or None, will download all runs. Defaults to None.
  • api (wandb.Api): The api object to use in requesting run information. If None or not provided, will create a new API access object using default values. Defaults to None.
  • logger (logger.Logger): The logger with which to report debug information. If None or not provided, no messages will be logged. Defaults to None.

Examples

Downloading all runs from sampottinger/who-wrote-this via the command line tool to a CSV file:

$ wandb-summarizer-to-csv sampottinger/who-wrote-this who_wrote_this.csv

Downloading select runs from sampottinger/who-wrote-this via the command line tool to a CSV file with no debug logging:

$ wandb-summarizer-to-csv sampottinger/who-wrote-this who_wrote_this.csv f "{\"config.corpusCol\": \"description\"}"

Downloading run information from sampottinger/who-wrote-this within another Python script:

import wandb_summarizer.download

run_info = wandb_summarizer.download.get_results('sampottinger/who-wrote-this')
print(run_info[0]['url'])

Purpose

This microlibrary builds summary datasets across all or many runs within a Weights and Biases project, reporting statistics from the end of each run as opposed to over time (over epochs) within those runs. This can be used to summarize inputs and results as a tabular data artifact across experiments within a project, supporting iterative model refinement or model selection. For example, users trying different dropout rates or L2 penalty inside a neural network for regularization may be logging those rates within Weights and Biases' config object (wandb.config). The wandb online interface may show those runs in a table with the final results from each but one may wish to visualize information from that table by building a bar chart showing the final validation / training set F1 scores at different dropout rates. As there is no "export table to CSV" option within wandb, one would need to write a custom script to download this data via Python. This microlibrary / command line tool makes it easy to download those results into a flat table for that post-hoc analysis, making it easier to generate results like the following:

L2 Penalty Final Validation Accuracy Final Training Accuracy
0.001 73% 87%
0.01 73% 75%
0.1 56% 55%

Development Standards

All top level methods should be unit tested and have Google style guide conformant docstrings. Please conform to the Google Python Style Guide when possible.


Testing

Automated tests are available and runnable via nosetests like so:

$ nosetests
.........
----------------------------------------------------------------------
Ran 9 tests in 1.331s

OK

Open Source Libraries Used

This uses the wandb client internally under the MIT License.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for wandb-summarizer, version 0.1.2
Filename, size File type Python version Upload date Hashes
Filename, size wandb_summarizer-0.1.2-py3-none-any.whl (9.1 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size wandb_summarizer-0.1.2.tar.gz (6.9 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page