A dot in the plot is a single observation (row in dataset) which means an instance type, optimization, and micro-architecture. In total (between test and train) there are about 5k experiments, and each experiment result is derived from 3 runs. These are instances on AWS. The Y-axis here is features, these are from node feature discovery. The Shapley value is how much the feature for the specific observation (experiment row) pushed the model's prediction away from the average prediction. When you see a "cat" prefix of a feature, it means it was categorical and one-hot encoded. Please see the Performance Cluster to understand how the features were normalized.
Calibrating telemetry...