Add chunk base level activations?

[asmaloney]

ACT-R provides ways to set base level activations on individual chunk types directly instead of computing it. This is used in the spreading activation equation (pp. 290-291).

vanilla

There does not seem to be a way to set the base level when declaring a chunk.

set-base-levels (pg. 326)

Explicitly sets a base level on the listed chunks.

This will set c's base level to 10:

(define-model test-base-levels
       (sgp :esc t :bll nil)
       (add-dm (a isa chunk)
               (b isa chunk)
               (c isa chunk)))

(set-base-levels (c 10))

set-all-base-levels (pg. 326)

Set it on all chunks.

(set-all-base-levels 1.5)

ccm

Base levels seem to be set using the DMBaseLevel memory sub-module, but I haven not yet found an example of how it works.

pyactr

These might be set using DecMem's add_activation? I will have to look at the math to see if that's what's happening. There are no examples using it.

In addition, there seems to be different defaults from ACT-R w.r.t. base-level learning.

The baselevel_learning function in the utilities takes bll and decay as parameters which are passed in from the model's parameters:

• baselevel_learning: a boolean which defaults to True
• decay: a float which defaults to 0.5

In ACT-R, however, :bll controls both:

The base-level learning parameter controls whether base-level learning is enabled, and also sets the value of the decay parameter, d. It can be set to any positive number or the value nil. The value nil means do not use base-level learning and is the default value, a number means that base-level is enabled and the given value is the decay parameter. The recommended value for :bll is .5, and it is one of the few parameters which have a strong recommended value. (pg. 298)

So to match ACT-R, the default in pyactr should be baselevel_learning=False.

[asmaloney]

So to match ACT-R, the default in pyactr should be baselevel_learning=False.

Note that right now this breaks things. See #391.