Dilation Operations

[dev0Guy]

Support Multiple Reduce Functions in MetricCluster to Preserve Operation Dimension

Problem

Currently, dilation is only defined for MetricCluster, and the cluster accepts only one operation (reduce) function.

Because of this limitation, the original observation space (number_of_machine, number_of_resources, n_ticks) is transformed into (machine_grid_x, machine_grid_y, number_of_resource, n_tick).

However, this is effectively a flattened version of (machine_grid_x, machine_grid_y, number_of_operation = 1, number_of_resource, n_tick). The number_of_operation dimension is implicitly fixed to 1, which restricts feature generation flexibility.

Desired Behavior

We would like MetricCluster to support multiple reduce functions so that it can generate n feature maps instead of just one.

The expected observation space should be (machine_grid_x, machine_grid_y, number_of_operation, number_of_resource, n_tick), where number_of_operation = n corresponds to the number of reduce functions provided.

Proposed Changes

1. Allow MetricCluster to accept a list of reduce functions instead of a single function.
2. Apply each reduce function independently.
3. Stack the results along a new number_of_operation dimension.

Resulting Output Shape

The output shape should change from (machine_grid_x, machine_grid_y, number_of_resource, n_tick) to (machine_grid_x, machine_grid_y, number_of_operation, number_of_resource, n_tick).