Dataset - Analytics and Charts

Once a DatasetVersion is annotated, Picsellia gives you two complementary ways to assess and explore it in aggregate, on top of browsing Asset one by one: the Analytics tab, and the Charts tab.

1. Analytics

Once your DatasetVersion is annotated, Picsellia automatically computes several metrics that can be used to assess the quality of your Dataset.

To visualize them, you can jump into the Analytics tab of your DatasetVersion:

You can obviously hover over any bar from any chart to get the analytics details.

2. Charts

Charts let you build custom aggregate visualizations over the Asset of a DatasetVersion, alongside the Query Language and Embeddings exploration modes detailed here.

For a Computer Vision project, this kind of aggregate view is what surfaces invisible problems Asset by Asset: a breakdown by Label can reveal a class imbalance before it skews training, a breakdown by Custom Metadata can help track representativity of the current DatasetVersion, and a breakdown by AssetTag can confirm a train/test/val split is balanced as expected.

A. Accessing Charts

Like the Datalake, Charts are available from the Charts tab at the top of the DatasetVersion, listing every chart created so far and from the side-by-side exploration mode combining Charts with the Assets overview.

Clicking on a bar of a chart automatically filters the Asset displayed in the Assets overview to only the ones matching that bar, by applying the equivalent query behind the scenes, the same way it does on the Datalake, detailed here.

B. Creating a Chart

Click + Create chart to open the creation form:

• Name: the name of your chart.
• Aggregate Function: AVG, COUNT, MAX, MIN, or SUM.
• Aggregate Field: the field to aggregate, for instance ID (to count Asset), DATA_CUSTOM_METADATA, SHAPE_ATTRIBUTES, or ANNOTATION_ATTRIBUTES (i.e. Asset Attribute).
• Group By Field: the field used to bucket the results. Asset-specific fields include CREATED_AT, TAG (i.e. AssetTag), SHAPE_LABEL, SHAPE_TEXT, SHAPE_AREA, SHAPE_ATTRIBUTES, and ANNOTATION_ATTRIBUTES (i.e. Asset Attribute). You can also group by fields inherited from the underlying Data: DATA_CREATED_AT, DATA_CONTENT_TYPE, DATA_UPLOAD_STATUS, DATA_FILE_SIZE, DATA_TAG, DATA_SOURCE, DATA_METADATA, DATA_CUSTOM_METADATA, DATA_DATALAKE, or DATA_DEPLOYMENT.
• Advanced filtering: optionally scope the chart to a subset of Asset using the same Query Language as the Search Bar, detailed here.

For instance, below a Chart showing the average Shape area grouped by Label name is being created:

Once created, your chart is displayed as a card showing the resulting visualization, for instance a bar chart of the average Shape area aggregated by SHAPE_LABEL.

Please note that a Chart can be created the same way from the side-by-side view:

C. Managing Charts

From a chart's card, you can:

• Recompute: refresh the chart with the latest Asset and Annotation.
• Update or Delete: from the ... menu.