How to use the N-dimensional travel distance workflow

The N-dimensional travel distance workflow is similar to the embedding space travel distance workflow. However, instead of using the 3D (or 2D) embedding values, this workflow can use any metrics (i.e., columns) to compute the travel distance in that N-dimensional space.

The N-dimensional travel distance workflow requires metrics collected over multiple epochs. The travel distance is computed from one epoch to the next and then is summed as total travel distance for each sample. The higher travel distance a sample has, the more challenging the sample could be to learn.

Below is an example of how to compute N-dimensional travel distance in the space of Loss, Confidence, and RMSE. Select the three columns, RIGHTCLICK on one of the column’s header, and then select N-dimensional travel distance under Workflows in the popup menu.

As in other workflows, you will be guided through the workflow with instructions in each step. You may want to skip the intermediate steps to reach the final step.

Once you get the travel distance, you may want to assign higher weights for those samples with high travel distances, so that they can be exposed more frequently during training.