Small fixes and new warning for Neurons classes.

[colleenjg]

Small fixes and new warning for Neurons classes.

Fixes:

• Added an error in VectorCells init if class is initialized directly.
• Removed default value of None for Other_Agent in AgentVectorCells init, as None is not an allowed value.
• Raised an error in ObjectVectorCells if there are no objects in the environment.

New warning added:

• Added a warning in classes where the number of neurons can be changed internally, overriding params["n"]. This is to avoid confusion when a user requests 10 neurons, but ends up with 64, for example. The added code specifically checks whether n was passed as a parameter, and will not produce a warning if n was added in from a default parameter dictionary.

New warning examples:

1. In the simple case that applies to HeadDirectionCells in 1D environments and to SpeedCell, if you run something like:

from ratinabox import Agent, Environment, SpeedCell

Env = Environment(params={"objects": [[0.5, 0.5]]})
Ag = Agent(Env)
SpeedCell(Ag, params={"n": 3})

you get:

UserWarning: Ignoring 'n' parameter value (3) that was passed for SpeedCell. Only 1 speed cell is needed.

2. In the slightly more complex case that applies to VectorCells and its child classes, if you run something like

from ratinabox import Agent, Environment, FieldOfViewOVCs

Env = Environment(params={"objects": [[0.5, 0.5]]})
Ag = Agent(Env)

FieldOfViewOVCs(Ag, params={"object_tuning_type": "random", "cell_arrangement": "diverging_manifold"})
FieldOfViewOVCs(Ag, params={"object_tuning_type": "random", "cell_arrangement": "random", "n":10})

then there's no warning. In the first case, because you didn't specify an n value [1] and in the second case, because the n value is not changed.

However, if you run

from ratinabox import Agent, Environment, FieldOfViewOVCs

Env = Environment(params={"objects": [[0.5, 0.5]]})
Ag = Agent(Env)
FieldOfViewOVCs(Ag, params={"object_tuning_type": "random", "cell_arrangement": "diverging_manifold", "n": 10})

you get:

UserWarning: Ignoring 'n' parameter value (10) that was passed, and setting number of ObjectVectorCell neurons to 58, inferred from the cell arrangement parameter.

[1] To ensure that no superfluous warning is produced in this case, I had to create an extra attribute _warn_n_change that can be set in a child class. This allows the init of VectorCells to know whether n was actively passed by the user or was added to params at some point during the initialization chain (e.g., from a child class' default parameters).

[TomGeorge1234]

As discussed in person these look great...and I'm happy with your solution to the awkward issue of the n parameter for vector cells, we can always refine it down the line but its only a tiny warning for good housekeeping after all

[TomGeorge1234]

Enormous thanks, as per, for your work on this!