What-If Tool
What-If Tool
Learn > Basics of the What-If Tool > How to - Customize the Datapoints Visualization

How to - Customize the Datapoints Visualization

The Datapoint Editor workspace has two types of visualizations in the playground: the Datapoints visualization and the Partial Dependence Plots visualization (discussed in a separate article).

The Datapoints visualization is the default visualization in the Datapoint Editor, and is built on Facets Dive. It can be used to arrange individual data points into visualizations customized by any features available in the data set. This way, users can investigate individual data points in the context of the loaded dataset.

To customize the Datapoints visualization:

  1. Select Datapoints in the Visualize module to see the datapoints visualization. The What-If Tool looks for meaningful defaults based on the model type and number of available models to help users get started.
  2. Using the dropdowns in the toolbar on top, specify features to visualize on the X-axis, Y-axis, color or labels. Combinations of these can be used to create histograms or scatterplots, or combinations of the two: a. Binning | X-Axis: This dropdown is used to select a feature to organize data points into buckets as a histogram on the X-Axis. The count dropdown specifies the number of buckets to create. b. Binning | Y-Axis: This dropdown is used to select a feature to organize data points into buckets as a histogram on the Y-Axis. The count dropdown specifies the number of buckets to create.

c. Color by: This dropdown is used to select a feature to color the data points. i. If a categorical feature is selected, a different color is used for each category. ii. If a numerical feature is selected, data points are colored using a two-color (sequential) colorscale.

d. Label by: This dropdown is used to specify a feature to label by. i. Numerical and categorical values are rendered as text over the Datapoint. ii. Features with image values are displayed as thumbnails

e. Scatter by | X-Axis: This dropdown is used to select a numerical feature to scatter data points on the X-Axis. The range of the X-Axis is determined by the minimum and maximum of selected feature values.

f. Scatter by | Y-Axis: This dropdown is used to select a numerical feature to scatter data points on the Y-Axis. The range of the X-Axis is determined by the minimum and maximum of selected feature values.

  1. Setting a dropdown value to none removes the features from the visualization.

Other interactions in the Datapoints visualization include:

  1. To select a datapoint, click on the datapoint and see individual feature values in the Edit module.
  2. To deselect a datapoint, click anywhere else in the visualization.
  3. To zoom into the visualization, click on the zoom in (+) icon or scroll in using your mouse.
  4. To zoom out of the visualization, click on the zoom out (-) icon or scroll out using your mouse.
  5. To reset the visualization, click on the frame icon.
  6. To pan the visualization, click and drag the mouse.
  7. Collapse or expand the legend by clicking on the down icon in the header.
Confusion Matrices and Small Multiples
You can create a variety of visualizations using different permutations and combinations of features for intersectional analysis quickly and iteratively. For binary classification models, easily create a confusion matrix by binning by the ground truth label on the X-axis and the predicted label on the Y-Axis. This should give four bins corresponding to false positives, true positives, false negatives, and true negatives. Easily convert these into small multiples visualizations by specifying features to scatter by on the X-Axis and Y-axis.
time to read
15 minutes
use with
Classification models
Multi-class models
Regression models
before you begin
N/A
related demos
Binary Classification Model: UCI Census Income Prediction

Multi-class Classification Model: Flowers Species Identification

Regression Model: UCI Census Age Prediction
takeaways
Use features in a loaded dataset to customize visualizations in the Datapoint Visualization.
what-if questions
Are there correlations between features and model predictions?

What differences between two data points cause models to behave differently?

Which data points on which do two models agree or disagree on?
The What-If Tool is being actively developed and documentation is likely to change as we improve the tool. We want to hear from you! Leave us a note, feedback, or suggestion on our community group.
Google Research
Get Involved
Github