Examples#

Exploring Image Datasets#

Run the example Jupyter notebook here.

Symphony has a set of widgets that use preprocessing to analyze large ML datasets. In this example, we look at the CIFAR-10 image classification dataset and use DNIKit to generate an interactive dataset report.

The example contains a precomptued analysis and demonstrates a handful of Symphony widgets:

  • Summary: A summary of the datset with distribution charts for each column in the dataset.

  • List : An browsable interface to explore the datset instances.

  • Scatterplot: An interactive embedding visualization to help with cluster identification.

  • Familiarity: The familiarity widget calculates how common each image is relative to the whole dataset. It can be used to find outliers and uncommon classes.

  • Duplicates: The duplicates widget calculates which images are the most visually similar or may be the same instance.

It can be used to find train/test overlap or other data redundancy issues.

Widget Examples#

Each widget contains an example. Check out each in the Available Components.