cambrian.ml.constraint_fns

These are constraint functions for the optimizers. These functions are used to prune experiments from the search space.

Functions

nevergrad_constraint_fn(parameterization, /, *, fn, ...)

This function is used to prune experiments for nevergrad sweepers. It will

constrain_total_pixels(*, num_eyes_to_generate, ...)

This constraint method will check whether the total number of pixels generated

constrain_total_memory_throughput(*, ...)

This constraint method will check whether the total number of pixels generated

constrain_morphologically_feasible_eyes(*, ...[, ...])

This constraint method will check whether the eye config, if placed

constrain_total_num_eyes(*, num_eyes_to_generate, ...)

This constraint method will check whether the total number of eyes generated

Module Contents

nevergrad_constraint_fn(parameterization, /, *, fn, **parameters)[source]

This function is used to prune experiments for nevergrad sweepers. It will return False if the experiment should be pruned.

constrain_total_pixels(*, num_eyes_to_generate, resolution, max_num_pixels)[source]

This constraint method will check whether the total number of pixels generated is less than a certain threshold.

constrain_total_memory_throughput(*, num_eyes_to_generate, resolution, stack_size, max_pixels_in_memory)[source]

This constraint method will check whether the total number of pixels generated is less than a certain threshold.

constrain_morphologically_feasible_eyes(*, num_eyes_to_generate, resolution, lon_range, radius=0.1, pixel_size=0.005, **_)[source]

This constraint method will check whether the eye config, if placed num_eyes_to_generate along the longitude of the agent, would be morphologically feasible. Morphologically feasible in this approximated case is basically whether all the eyes would fit. There are two primary factors here:

1. sensorsize and number of eyes. We want to make sure, along the horizontal axis, that the eyes don’t overlap.

2. The total number of pixels. We want to make sure that the total number of pixels generated is less than a certain threshold.

Going to approximate the agent as a circle and the eyes as a line with a length equal to the sensorsize width. Then we’ll check whether the eyes fit in the allowed longitude range.

Parameters:

  • num_eyes_to_generate (int) – The number of eyes to generate along the longitude of the agent.

  • resolution (Tuple[int, int] | int) – The resolution of the eye.

  • lon_range (Tuple[int, int] | int) – The range of longitudes in which to generate the eyes. This is in degrees.

Keyword Arguments:

  • radius (float) – The radius of the agent. Default is 0.2.

  • pixel_size (float) – The pixel size of the eye. This is used to calculate the total width of the eyes. Default is 0.01.

constrain_total_num_eyes(*, num_eyes_to_generate, max_num_eyes)[source]

This constraint method will check whether the total number of eyes generated is less than a certain threshold.