:orphan:

alpha_w
*******

**Note:** Only applicable in the A-learning and the Actor-Critic mechanisms.

Specifies the learning rate for w-values (conditioned reinforcement for a stimulus element),
used in the updating of w-values. See :ref:`the-mechanisms`.

The parameter ``alpha_w`` can be set to certain values for certain particular stimulus elements.
It can also be specified as a single value, in which case the same learning rate is used for all
stimulus elements. Wildcards (*) may be used to set several values at once.

Syntax
------

::

  alpha_w = e1: v1, e2: v2, ..., en: vn, default: d
  alpha_w = v1
  alpha_w = *:v1  # Same as above

where ``v1,v2,...,vn`` and ``d`` are :ref:`scalar expressions<scalar-expressions>`.

Description
-----------

``alpha_w = e1: v1, e2: v2, ..., en->bn: vn, default: d`` sets the learning rate for

- w(e1) to v1,
- w(e2) to v2, ...,
- w(en) to vn,

and the learning rate for all other w-values to d.

- The specification is independent of the list order:

  ``alpha_w = e1:v1, e2:v2, default:d``

  is the same as

  ``alpha_w = e2:v2, default:d, e1:v1``.

- ``default`` need not be specified if all stimulus elements are present in the list. For example,

  ::

    stimulus_elements = e1, e2
    alpha_w = e1:v1, e2:v2

----

``alpha_w = v1`` sets the learning rate for all w-values to v1.

- ``alpha_w = v1`` is the same as ``alpha_w = default: v1``.

Dependencies
------------

- The property ``stimulus_elements`` must be specified before ``alpha_w``.
- Each stimulus element used in the specification of ``alpha_w`` must be present in the parameter ``stimulus_elements``.
- Applicable only in the A-learning and Actor-Critic mechanisms.

Examples
--------

::

  @variables x = 1
  alpha_w = element1: x, element2: x+1, default:x+2

sets the learning rate for w(element1) to 1 and for w(element2) to 2, and
for the remaining stimulus elements to 3.

::

  alpha_w = element1: 0.5, default:0

sets the learning rate for w(element1) to 0.5, and for the remaining stimulus elements to 3.

::

  alpha_w = 0.1

sets the learning rate for all w-values to 0.1.