11972746306

Committed 22 Nov 2024 12:39PM UTC coverage: 83.896% (+0.08%) from 83.818%

Build # 11972746306

Build Type

push

github

Committed by

web-flow

Commit Message

refactor/emcomposition_field_handling (#3122)

* • emcomposition.py
  _parse_fields:  clean up assignment of self.num_fields

• test_emcomposition.py
  add test_assign_field_weights_and_0_vs_None()
  add test_field_weights_all_None_and_or_0

• emcomposition.py
  - revamp docstring to document new mods
  - add fields arg to specify field_naes, field_weights, learn_field_weights
  - implement fields arg to specify field_names, field_weights, learn_field_weights in dict format
  - implement support for field-specific learn_field_weight specifications
  - _identify_target_nodes():
    refactor to use target_fields instead of learn_field_weights
  - add target_fields to fields specification dict
  - add dict spec for entries in fields arg
  - start adding field_idx to all components
  - add self._field_index_map

• pytorchEMcompositionwrapper.py
  - store_memory(): use self._field_index_map to assign memories to fields

• test_emcomposition.py
  - test_backpropagation_of_error_in_learning(): use EGO model to test for error backpropagation through EMCompoistion
  - test_field_args_and_map_assignments(): flesh out _field_index_map validation

* • emcomposition.py
  - update docstring figs
  - add purge_by_field_weights Parameter

* • autodiffcomposition.py
  - infer_backpropagation_learning_pathways(): add NodeRole.BIAS to pathways consructed for learning

Run Details

9458 of 12496 branches covered (75.69%)

Branch coverage included in aggregate %.

141 of 149 new or added lines in 4 files covered. (94.63%)

6 existing lines in 2 files now uncovered.

32778 of 37847 relevant lines covered (86.61%)

0.87 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

82.79

/psyneulink/library/components/projections/pathway/autoassociativeprojection.py

# Princeton University licenses this file to You under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.  You may obtain a copy of the License at:
#     http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software distributed under the License is distributed
# on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and limitations under the License.


# *******************************************  AutoAssociativeProjection ***********************************************

"""

Contents
--------

  * `AutoAssociative_Overview`
  * `AutoAssociative_Creation`
  * `AutoAssociative_Structure`
      - `AutoAssociative_Configurable_Attributes`
  * `AutoAssociative_Execution`
  * `AutoAssociative_Class_Reference`


.. _AutoAssociative_Overview:

Overview
--------

An AutoAssociativeProjection is a subclass of `MappingProjection` that acts as the recurrent projection for a
`RecurrentTransferMechanism`. The primary difference between an AutoAssociativeProjection and a basic MappingProjection
is that an AutoAssociativeProjection uses the `auto <RecurrentTransferMechanism.auto>` and
`hetero <RecurrentTransferMechanism.hetero>` parameters *on the RecurrentTransferMechanism* to help update its matrix:
this allows for a `ControlMechanism <ControlMechanism>` to control the `auto <RecurrentTransferMechanism.auto>` and
`hetero <RecurrentTransferMechanism.hetero>` parameters and thereby control the matrix.

AutoAssociativeProjection represents connections between nodes in a single-layer recurrent network. It multiplies
the output of the `RecurrentTransferMechanism` by a matrix, then presents the product as input to the
`RecurrentTransferMechanism`.



.. _AutoAssociative_Creation:

Creating an AutoAssociativeProjection
-------------------------------------

An AutoAssociativeProjection is created automatically by a RecurrentTransferMechanism (or its subclasses), and is
stored as the `recurrent_projection <RecurrentTransferMechanism.recurrent_projection>` parameter of the
RecurrentTransferMechanism. It is not recommended to create an AutoAssociativeProjection on its own, because during
execution an AutoAssociativeProjection references parameters owned by its RecurrentTransferMechanism (see
`Execution <AutoAssociative_Execution>` below).

.. _AutoAssociative_Structure:

Auto Associative Structure
--------------------------

In structure, the AutoAssociativeProjection is almost identical to a MappingProjection: the only additional attributes
are `auto <AutoAssociativeProjection.auto>` and `hetero <AutoAssociativeProjection.hetero>`.

.. _AutoAssociative_Configurable_Attributes:

*Configurable Attributes*
~~~~~~~~~~~~~~~~~~~~~~~~~

Due to its specialized nature, most parameters of the AutoAssociativeProjection are not configurable: the `variable` is
determined by the format of the output of the RecurrentTransferMechanism, the `function` is always MatrixTransform, and so
on. The only configurable parameter is the matrix, configured through the **matrix**, **auto**, and/or **hetero**
arguments for a RecurrentTransferMechanism:

.. _AutoAssociative_Matrix:

* **matrix** - multiplied by the input to the AutoAssociativeProjection in order to produce the output. Specification of
  the **matrix**, **auto**, and/or **hetero** arguments determines the values of the matrix; **auto** determines the
  diagonal entries (representing the strength of the connection from each node to itself) and **hetero** determines
  the off-diagonal entries (representing connections between nodes).

.. _AutoAssociative_Execution:

Execution
---------

An AutoAssociativeProjection uses its `matrix <AutoAssociativeProjection.matrix>` parameter to transform the value of
its `sender <AutoAssociativeProjection.sender>`, and provide the result as input for its
`receiver <AutoAssociativeProjection.receiver>`, the primary InputPort of the RecurrentTransferMechanism.

.. note::
     During execution the AutoAssociativeProjection updates its `matrix <AutoAssociativeProjection.matrix> parameter
     based on the `auto <RecurrentTransferMechanism.auto>` and `hetero <RecurrentTransferMechanism.hetero>` parameters
     *on the `RecurrentTransferMechanism`*. (The `auto <AutoAssociativeProjection.auto>` and
     `hetero <AutoAssociativeProjection.hetero>` parameters of the AutoAssociativeProjection simply refer to their
     counterparts on the RecurrentTransferMechanism as well.) The reason for putting the `auto
     <RecurrentTransferMechanism.auto>` and `hetero <RecurrentTransferMechanism.hetero>` parameters on the
     RecurrentTransferMechanism is that this allows them to be modified by a `ControlMechanism <ControlMechanism>`.

.. _AutoAssociative_Class_Reference:

Class Reference
---------------

"""
import numpy as np
from beartype import beartype

from psyneulink._typing import Optional

from psyneulink.core.components.component import parameter_keywords
from psyneulink.core.components.functions.nonstateful.transformfunctions import MatrixTransform
from psyneulink.core.components.functions.function import get_matrix
from psyneulink.core.components.projections.pathway.mappingprojection import MappingError, MappingProjection
from psyneulink.library.components.projections.pathway.maskedmappingprojection import MaskedMappingProjection
from psyneulink.core.components.projections.projection import projection_keywords
from psyneulink.core.components.shellclasses import Mechanism
from psyneulink.core.components.ports.outputport import OutputPort
from psyneulink.core.globals.keywords import AUTO_ASSOCIATIVE_PROJECTION, DEFAULT_MATRIX, HOLLOW_MATRIX, FUNCTION, OWNER_MECH
from psyneulink.core.globals.parameters import SharedParameter, Parameter, check_user_specified
from psyneulink.core.globals.preferences.basepreferenceset import ValidPrefSet
from psyneulink.core.globals.preferences.preferenceset import PreferenceLevel
from psyneulink.core.globals.utilities import is_numeric_scalar

__all__ = [
    'AutoAssociativeError', 'AutoAssociativeProjection', 'get_auto_matrix', 'get_hetero_matrix',
]

parameter_keywords.update({AUTO_ASSOCIATIVE_PROJECTION})
projection_keywords.update({AUTO_ASSOCIATIVE_PROJECTION})


class AutoAssociativeError(MappingError):
    pass


class AutoAssociativeProjection(MaskedMappingProjection):
    """
    AutoAssociativeProjection(
        )

    Subclass of `MappingProjection` that is self-recurrent on a `RecurrentTransferMechanism`.
    See `MappingProjection <MappingProjection_Class_Reference>` and `Projection <Projection_Class_Reference>`
    for additional arguments and attributes.

    COMMENT:
        JDC [IN GENERAL WE HAVE TRIED TO DISCOURAGE SUCH DEPENDENCIES;  BETTER TO HAVE IT ACCEPT ARGUMENTS THAT
        RecurrentTransferMechanism (or Composition) PROVIDES THAN ASSUME THEY ARE ON ANOTHER OBJECT THAT CREATED
        THIS ONE]
    Note: The reason **auto** and **hetero** are not arguments to the constructor of the AutoAssociativeProjection is
    because it is only ever created by a RecurrentTransferMechanism: by the time the AutoAssociativeProjection is
    created, the **auto** and **hetero** arguments are already incorporated into the **matrix** argument.

    COMMENT

    Arguments
    ---------

    sender : OutputPort or Mechanism : default None
        specifies the source of the Projection's input; must be (or belong to) the same Mechanism as **receiver**,
        and the length of its `value <OutputPort.value>` must match that of the `variable <InputPort.variable>` of
        the **receiver**.

    receiver: InputPort or Mechanism : default None
        specifies the destination of the Projection's output; must be (or belong to) the same Mechanism as **sender**,
        and the length of its `variable <InputPort.variable>` must match the `value <OutputPort.value>` of **sender**.

    matrix : list, np.ndarray, function or keyword : default DEFAULT_MATRIX
        specifies the matrix used by `function <Projection_Base.function>` (default: `LinearCombination`) to
        transform the `value <Projection_Base.value>` of the `sender <MappingProjection.sender>` into a value
        provided to the `variable <InputPort.variable>` of the `receiver <MappingProjection.receiver>` `InputPort`;
        must be a square matrix (i.e., have the same number of rows and columns).

    Attributes
    ----------

    sender : OutputPort
        the `OutputPort` of the `Mechanism <Mechanism>` that is the source of the Projection's input; in the case of
        an AutoAssociativeProjection, it is an OutputPort of the same Mechanism to which the `receiver
        <AutoAssociativeProjection.receiver>` belongs.

    receiver: InputPort
        the `InputPort` of the `Mechanism <Mechanism>` that is the destination of the Projection's output; in the case
        of an AutoAssociativeProjection, it is an InputPort of the same Mechanism to which the `sender
        <AutoAssociativeProjection.sender>` belongs.

    matrix : 2d np.ndarray
        square matrix used by `function <AutoAssociativeProjection.function>` to transform input from the `sender
        <MappingProjection.sender>` to the value provided to the `receiver <AutoAssociativeProjection.receiver>`;
        in the case of an AutoAssociativeProjection.

    """

    componentType = AUTO_ASSOCIATIVE_PROJECTION
    className = componentType
    suffix = " " + className

    class Parameters(MappingProjection.Parameters):
        """
            Attributes
            ----------

                variable
                    see `variable <AutoAssociativeProjection.variable>`

                    :default value: numpy.array([[0]])
                    :type: ``numpy.ndarray``
                    :read only: True

                auto
                    see `auto <AutoAssociativeProjection.auto>`

                    :default value: 1
                    :type: ``int``

                function
                    see `function <AutoAssociativeProjection.function>`

                    :default value: `MatrixTransform`
                    :type: `Function`

                hetero
                    see `hetero <AutoAssociativeProjection.hetero>`

                    :default value: 0
                    :type: ``int``

                matrix
                    see `matrix <AutoAssociativeProjection.matrix>`

                    :default value: `AUTO_ASSIGN_MATRIX`
                    :type: ``str``
        """
        variable = Parameter(np.array([[0]]), read_only=True, pnl_internal=True, constructor_argument='default_variable')
        # function is always MatrixTransform that requires 1D input
        function = Parameter(MatrixTransform, stateful=False, loggable=False)

        auto = SharedParameter(1, attribute_name=OWNER_MECH)
        hetero = SharedParameter(0, attribute_name=OWNER_MECH)
        matrix = SharedParameter(DEFAULT_MATRIX, attribute_name=OWNER_MECH)

    classPreferenceLevel = PreferenceLevel.TYPE

    @check_user_specified
    @beartype
    def __init__(self,
                 owner=None,
                 sender=None,
                 receiver=None,
                 matrix=None,
                 function=None,
                 params=None,
                 name=None,
                 prefs:  Optional[ValidPrefSet] = None,
                 **kwargs
                 ):

        if owner is not None:
            if not isinstance(owner, Mechanism):
                raise AutoAssociativeError('Owner of AutoAssociative Mechanism must either be None or a Mechanism')
            if sender is None:
                sender = owner
            if receiver is None:
                receiver = owner

        super().__init__(sender=sender,
                         receiver=receiver,
                         matrix=matrix,
                         function=function,
                         params=params,
                         name=name,
                         prefs=prefs,
                         **kwargs)

    # temporary override to make sure matrix/auto/hetero parameters
    # get passed properly. should be replaced with a better organization
    # of auto/hetero, in which the base parameters are stored either on
    # AutoAssociativeProjection or on MatrixTransform itself
    def _instantiate_parameter_classes(self, context):
        if FUNCTION not in self.initial_shared_parameters:
            try:
                self.initial_shared_parameters[FUNCTION] = {
                    'matrix': self.initial_shared_parameters[OWNER_MECH]['matrix']
                }
            except KeyError:
                pass

        super()._instantiate_parameter_classes(context)

    # COMMENTED OUT BY KAM 1/9/2018 -- this method is not currently used; should be moved to Recurrent Transfer Mech
    #     if it is used in the future

    # def _update_auto_and_hetero(self, owner_mech=None, runtime_params=None, time_scale=TimeScale.TRIAL, context=None):
    #     if owner_mech is None:
    #         if isinstance(self.sender, OutputPort):
    #             owner_mech = self.sender.owner
    #         elif isinstance(self.sender, Mechanism):
    #             owner_mech = self.sender
    #         else:
    #             raise AutoAssociativeError("The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently"
    #                                        " the sender is {}".
    #                                        format(self.__class__.__name__, self.name, self.sender))
    #     if AUTO in owner_mech._parameter_ports and HETERO in owner_mech._parameter_ports:
    #         owner_mech._parameter_ports[AUTO].update(context=context, params=runtime_params, time_scale=time_scale)
    #         owner_mech._parameter_ports[HETERO].update(context=context, params=runtime_params, time_scale=time_scale)
    #

    # END OF COMMENTED OUT BY KAM 1/9/2018

    # NOTE 7/25/17 CW: Originally, this override was written because if the user set the 'auto' parameter on the
        # recurrent mechanism, the ParameterPort wouldn't update until after the mechanism executed: since the system
        # first runs the projection, then runs the mechanism, the projection initially uses the 'old' value. However,
        # this is commented out because this may in fact be the desired behavior.
        # Two possible solutions: allow control to be done on projections, or build a more general way to allow
        # projections to read parameters from mechanisms.
    # def _update_parameter_ports(self, runtime_params=None, context=None):
    #     """Update this projection's owner mechanism's `auto` and `hetero` parameter ports as well! The owner mechanism
    #     should be a RecurrentTransferMechanism, which DOES NOT update its own `auto` and `hetero` parameter ports during
    #     its _update_parameter_ports function (so that the ParameterPort is not redundantly updated).
    #     Thus, if you want to have an AutoAssociativeProjection on a mechanism that's not a RecurrentTransferMechanism,
    #     your mechanism must similarly exclude `auto` and `hetero` from updating.
    #     """
    #     super()._update_parameter_ports(runtime_params, context)
    #
    #     if isinstance(self.sender, OutputPort):
    #         owner_mech = self.sender.owner
    #     elif isinstance(self.sender, Mechanism):
    #         owner_mech = self.sender
    #     else:
    #         raise AutoAssociativeError("The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently the"
    #                                    " sender is {}".
    #                                    format(self.__class__.__name__, self.name, self.sender))
    #
    #     if AUTO in owner_mech._parameter_ports and HETERO in owner_mech._parameter_ports:
    #         owner_mech._parameter_ports[AUTO].update(context=context, params=runtime_params)
    #         owner_mech._parameter_ports[HETERO].update(context=context, params=runtime_params)
    #     else:
    #         raise AutoAssociativeError("Auto or Hetero ParameterPort not found in {0} \"{1}\"; here are names of the "
    #                                    "current ParameterPorts for {1}: {2}".format(owner_mech.__class__.__name__,
    #                                    owner_mech.name, owner_mech._parameter_ports.key_values))

    @property
    def owner_mech(self):
        if isinstance(self.sender, OutputPort):
            return self.sender.owner
        elif isinstance(self.sender, Mechanism):
            return self.sender
        else:
            raise AutoAssociativeError(
                "The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently the sender is {}".format(
                    self.__class__.__name__, self.name, self.sender
                )
            )

    @property
    def matrix(self):
        owner_mech = self.owner_mech
        if hasattr(owner_mech, "matrix"):
            return owner_mech.matrix
        return super(AutoAssociativeProjection, self.__class__).matrix.fget(self)

    @matrix.setter
    def matrix(self, setting):
        owner_mech = self.owner_mech
        if hasattr(owner_mech, "matrix"):
            owner_mech.matrix = setting
        else:
            super(AutoAssociativeProjection, self.__class__).matrix.fset(self, setting)


# a helper function that takes a specification of `hetero` and returns a hollow matrix with the right values
def get_hetero_matrix(raw_hetero, size):
    if is_numeric_scalar(raw_hetero):
        return get_matrix(HOLLOW_MATRIX, size, size) * raw_hetero
    elif ((isinstance(raw_hetero, np.ndarray) and raw_hetero.ndim == 1) or
              (isinstance(raw_hetero, list) and np.array(raw_hetero).ndim == 1)):
        if len(raw_hetero) != 1:
            return None
        return get_matrix(HOLLOW_MATRIX, size, size) * raw_hetero[0]
    elif (isinstance(raw_hetero, np.matrix) or
              (isinstance(raw_hetero, np.ndarray) and raw_hetero.ndim == 2) or
              (isinstance(raw_hetero, list) and np.array(raw_hetero).ndim == 2)):
        np.fill_diagonal(raw_hetero, 0)
        return np.array(raw_hetero)
    else:
        return None


# similar to get_hetero_matrix() above
def get_auto_matrix(raw_auto, size):
    if is_numeric_scalar(raw_auto):
        return np.diag(np.full(size, raw_auto, dtype=float))
    elif ((isinstance(raw_auto, np.ndarray) and raw_auto.ndim == 1) or
              (isinstance(raw_auto, list) and np.array(raw_auto).ndim == 1)):
        if len(raw_auto) == 1:
            return np.diag(np.full(size, raw_auto[0], dtype=float))
        else:
            if len(raw_auto) != size:
                return None
            return np.diag(raw_auto)
    elif (isinstance(raw_auto, np.matrix) or
              (isinstance(raw_auto, np.ndarray) and raw_auto.ndim == 2) or
              (isinstance(raw_auto, list) and np.array(raw_auto).ndim == 2)):
        # we COULD add a validation here to ensure raw_auto is diagonal, but it would slow stuff down.
        return np.array(raw_auto)
    else:
        return None

1	# Princeton University licenses this file to You under the Apache License, Version 2.0 (the "License");
2	# you may not use this file except in compliance with the License. You may obtain a copy of the License at:
3	# http://www.apache.org/licenses/LICENSE-2.0
4	# Unless required by applicable law or agreed to in writing, software distributed under the License is distributed
5	# on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
6	# See the License for the specific language governing permissions and limitations under the License.
7
8
9	# ***************************************** AutoAssociativeProjection *********************************************
10
11	"""
12
13	Contents
14	--------
15
16	* `AutoAssociative_Overview`
17	* `AutoAssociative_Creation`
18	* `AutoAssociative_Structure`
19	- `AutoAssociative_Configurable_Attributes`
20	* `AutoAssociative_Execution`
21	* `AutoAssociative_Class_Reference`
22
23
24	.. _AutoAssociative_Overview:
25
26	Overview
27	--------
28
29	An AutoAssociativeProjection is a subclass of `MappingProjection` that acts as the recurrent projection for a
30	`RecurrentTransferMechanism`. The primary difference between an AutoAssociativeProjection and a basic MappingProjection
31	is that an AutoAssociativeProjection uses the `auto <RecurrentTransferMechanism.auto>` and
32	`hetero <RecurrentTransferMechanism.hetero>` parameters on the RecurrentTransferMechanism to help update its matrix:
33	this allows for a `ControlMechanism <ControlMechanism>` to control the `auto <RecurrentTransferMechanism.auto>` and
34	`hetero <RecurrentTransferMechanism.hetero>` parameters and thereby control the matrix.
35
36	AutoAssociativeProjection represents connections between nodes in a single-layer recurrent network. It multiplies
37	the output of the `RecurrentTransferMechanism` by a matrix, then presents the product as input to the
38	`RecurrentTransferMechanism`.
39
40
41
42	.. _AutoAssociative_Creation:
43
44	Creating an AutoAssociativeProjection
45	-------------------------------------
46
47	An AutoAssociativeProjection is created automatically by a RecurrentTransferMechanism (or its subclasses), and is
48	stored as the `recurrent_projection <RecurrentTransferMechanism.recurrent_projection>` parameter of the
49	RecurrentTransferMechanism. It is not recommended to create an AutoAssociativeProjection on its own, because during
50	execution an AutoAssociativeProjection references parameters owned by its RecurrentTransferMechanism (see
51	`Execution <AutoAssociative_Execution>` below).
52
53	.. _AutoAssociative_Structure:
54
55	Auto Associative Structure
56	--------------------------
57
58	In structure, the AutoAssociativeProjection is almost identical to a MappingProjection: the only additional attributes
59	are `auto <AutoAssociativeProjection.auto>` and `hetero <AutoAssociativeProjection.hetero>`.
60
61	.. _AutoAssociative_Configurable_Attributes:
62
63	Configurable Attributes
64	~~~~~~~~~~~~~~~~~~~~~~~~~
65
66	Due to its specialized nature, most parameters of the AutoAssociativeProjection are not configurable: the `variable` is
67	determined by the format of the output of the RecurrentTransferMechanism, the `function` is always MatrixTransform, and so
68	on. The only configurable parameter is the matrix, configured through the matrix, auto, and/or hetero
69	arguments for a RecurrentTransferMechanism:
70
71	.. _AutoAssociative_Matrix:
72
73	* matrix - multiplied by the input to the AutoAssociativeProjection in order to produce the output. Specification of
74	the matrix, auto, and/or hetero arguments determines the values of the matrix; auto determines the
75	diagonal entries (representing the strength of the connection from each node to itself) and hetero determines
76	the off-diagonal entries (representing connections between nodes).
77
78	.. _AutoAssociative_Execution:
79
80	Execution
81	---------
82
83	An AutoAssociativeProjection uses its `matrix <AutoAssociativeProjection.matrix>` parameter to transform the value of
84	its `sender <AutoAssociativeProjection.sender>`, and provide the result as input for its
85	`receiver <AutoAssociativeProjection.receiver>`, the primary InputPort of the RecurrentTransferMechanism.
86
87	.. note::
88	During execution the AutoAssociativeProjection updates its `matrix <AutoAssociativeProjection.matrix> parameter
89	based on the `auto <RecurrentTransferMechanism.auto>` and `hetero <RecurrentTransferMechanism.hetero>` parameters
90	on the `RecurrentTransferMechanism`. (The `auto <AutoAssociativeProjection.auto>` and
91	`hetero <AutoAssociativeProjection.hetero>` parameters of the AutoAssociativeProjection simply refer to their
92	counterparts on the RecurrentTransferMechanism as well.) The reason for putting the `auto
93	<RecurrentTransferMechanism.auto>` and `hetero <RecurrentTransferMechanism.hetero>` parameters on the
94	RecurrentTransferMechanism is that this allows them to be modified by a `ControlMechanism <ControlMechanism>`.
95
96	.. _AutoAssociative_Class_Reference:
97
98	Class Reference
99	---------------
100
101	"""
102	import numpy as np	1✔
103	from beartype import beartype	1✔
104
105	from psyneulink._typing import Optional	1✔
106
107	from psyneulink.core.components.component import parameter_keywords	1✔
108	from psyneulink.core.components.functions.nonstateful.transformfunctions import MatrixTransform	1✔
109	from psyneulink.core.components.functions.function import get_matrix	1✔
110	from psyneulink.core.components.projections.pathway.mappingprojection import MappingError, MappingProjection	1✔
111	from psyneulink.library.components.projections.pathway.maskedmappingprojection import MaskedMappingProjection	1✔
112	from psyneulink.core.components.projections.projection import projection_keywords	1✔
113	from psyneulink.core.components.shellclasses import Mechanism	1✔
114	from psyneulink.core.components.ports.outputport import OutputPort	1✔
115	from psyneulink.core.globals.keywords import AUTO_ASSOCIATIVE_PROJECTION, DEFAULT_MATRIX, HOLLOW_MATRIX, FUNCTION, OWNER_MECH	1✔
116	from psyneulink.core.globals.parameters import SharedParameter, Parameter, check_user_specified	1✔
117	from psyneulink.core.globals.preferences.basepreferenceset import ValidPrefSet	1✔
118	from psyneulink.core.globals.preferences.preferenceset import PreferenceLevel	1✔
119	from psyneulink.core.globals.utilities import is_numeric_scalar	1✔
120
121	__all__ = [	1✔
122	'AutoAssociativeError', 'AutoAssociativeProjection', 'get_auto_matrix', 'get_hetero_matrix',
123	]
124
125	parameter_keywords.update({AUTO_ASSOCIATIVE_PROJECTION})	1✔
126	projection_keywords.update({AUTO_ASSOCIATIVE_PROJECTION})	1✔
127
128
129	class AutoAssociativeError(MappingError):	1✔
130	pass	1✔
131
132
133	class AutoAssociativeProjection(MaskedMappingProjection):	1✔
134	"""
135	AutoAssociativeProjection(
136	)
137
138	Subclass of `MappingProjection` that is self-recurrent on a `RecurrentTransferMechanism`.
139	See `MappingProjection <MappingProjection_Class_Reference>` and `Projection <Projection_Class_Reference>`
140	for additional arguments and attributes.
141
142	COMMENT:
143	JDC [IN GENERAL WE HAVE TRIED TO DISCOURAGE SUCH DEPENDENCIES; BETTER TO HAVE IT ACCEPT ARGUMENTS THAT
144	RecurrentTransferMechanism (or Composition) PROVIDES THAN ASSUME THEY ARE ON ANOTHER OBJECT THAT CREATED
145	THIS ONE]
146	Note: The reason auto and hetero are not arguments to the constructor of the AutoAssociativeProjection is
147	because it is only ever created by a RecurrentTransferMechanism: by the time the AutoAssociativeProjection is
148	created, the auto and hetero arguments are already incorporated into the matrix argument.
149
150	COMMENT
151
152	Arguments
153	---------
154
155	sender : OutputPort or Mechanism : default None
156	specifies the source of the Projection's input; must be (or belong to) the same Mechanism as receiver,
157	and the length of its `value <OutputPort.value>` must match that of the `variable <InputPort.variable>` of
158	the receiver.
159
160	receiver: InputPort or Mechanism : default None
161	specifies the destination of the Projection's output; must be (or belong to) the same Mechanism as sender,
162	and the length of its `variable <InputPort.variable>` must match the `value <OutputPort.value>` of sender.
163
164	matrix : list, np.ndarray, function or keyword : default DEFAULT_MATRIX
165	specifies the matrix used by `function <Projection_Base.function>` (default: `LinearCombination`) to
166	transform the `value <Projection_Base.value>` of the `sender <MappingProjection.sender>` into a value
167	provided to the `variable <InputPort.variable>` of the `receiver <MappingProjection.receiver>` `InputPort`;
168	must be a square matrix (i.e., have the same number of rows and columns).
169
170	Attributes
171	----------
172
173	sender : OutputPort
174	the `OutputPort` of the `Mechanism <Mechanism>` that is the source of the Projection's input; in the case of
175	an AutoAssociativeProjection, it is an OutputPort of the same Mechanism to which the `receiver
176	<AutoAssociativeProjection.receiver>` belongs.
177
178	receiver: InputPort
179	the `InputPort` of the `Mechanism <Mechanism>` that is the destination of the Projection's output; in the case
180	of an AutoAssociativeProjection, it is an InputPort of the same Mechanism to which the `sender
181	<AutoAssociativeProjection.sender>` belongs.
182
183	matrix : 2d np.ndarray
184	square matrix used by `function <AutoAssociativeProjection.function>` to transform input from the `sender
185	<MappingProjection.sender>` to the value provided to the `receiver <AutoAssociativeProjection.receiver>`;
186	in the case of an AutoAssociativeProjection.
187
188	"""
189
190	componentType = AUTO_ASSOCIATIVE_PROJECTION	1✔
191	className = componentType	1✔
192	suffix = " " + className	1✔
193
194	class Parameters(MappingProjection.Parameters):	1✔
195	"""
196	Attributes
197	----------
198
199	variable
200	see `variable <AutoAssociativeProjection.variable>`
201
202	:default value: numpy.array([[0]])
203	:type: ``numpy.ndarray``
204	:read only: True
205
206	auto
207	see `auto <AutoAssociativeProjection.auto>`
208
209	:default value: 1
210	:type: ``int``
211
212	function
213	see `function <AutoAssociativeProjection.function>`
214
215	:default value: `MatrixTransform`
216	:type: `Function`
217
218	hetero
219	see `hetero <AutoAssociativeProjection.hetero>`
220
221	:default value: 0
222	:type: ``int``
223
224	matrix
225	see `matrix <AutoAssociativeProjection.matrix>`
226
227	:default value: `AUTO_ASSIGN_MATRIX`
228	:type: ``str``
229	"""
230	variable = Parameter(np.array([[0]]), read_only=True, pnl_internal=True, constructor_argument='default_variable')	1✔
231	# function is always MatrixTransform that requires 1D input
232	function = Parameter(MatrixTransform, stateful=False, loggable=False)	1✔
233
234	auto = SharedParameter(1, attribute_name=OWNER_MECH)	1✔
235	hetero = SharedParameter(0, attribute_name=OWNER_MECH)	1✔
236	matrix = SharedParameter(DEFAULT_MATRIX, attribute_name=OWNER_MECH)	1✔
237
238	classPreferenceLevel = PreferenceLevel.TYPE	1✔
239
240	@check_user_specified	1✔
241	@beartype	1✔
242	def __init__(self,	1✔
243	owner=None,
244	sender=None,
245	receiver=None,
246	matrix=None,
247	function=None,
248	params=None,
249	name=None,
250	prefs: Optional[ValidPrefSet] = None,
251	**kwargs
252	):
253
254	if owner is not None:	1!
255	if not isinstance(owner, Mechanism):	1✔
256	raise AutoAssociativeError('Owner of AutoAssociative Mechanism must either be None or a Mechanism')
257	if sender is None:	1✔
258	sender = owner	1✔
259	if receiver is None:	1✔
260	receiver = owner	1✔
261
262	super().__init__(sender=sender,	1✔
263	receiver=receiver,
264	matrix=matrix,
265	function=function,
266	params=params,
267	name=name,
268	prefs=prefs,
269	**kwargs)
270
271	# temporary override to make sure matrix/auto/hetero parameters
272	# get passed properly. should be replaced with a better organization
273	# of auto/hetero, in which the base parameters are stored either on
274	# AutoAssociativeProjection or on MatrixTransform itself
275	def _instantiate_parameter_classes(self, context):	1✔
276	if FUNCTION not in self.initial_shared_parameters:	1!
277	try:	1✔
278	self.initial_shared_parameters[FUNCTION] = {	1✔
279	'matrix': self.initial_shared_parameters[OWNER_MECH]['matrix']
280	}
281	except KeyError:	×
282	pass	×
283
284	super()._instantiate_parameter_classes(context)	1✔
285
286	# COMMENTED OUT BY KAM 1/9/2018 -- this method is not currently used; should be moved to Recurrent Transfer Mech
287	# if it is used in the future
288
289	# def _update_auto_and_hetero(self, owner_mech=None, runtime_params=None, time_scale=TimeScale.TRIAL, context=None):
290	# if owner_mech is None:
291	# if isinstance(self.sender, OutputPort):
292	# owner_mech = self.sender.owner
293	# elif isinstance(self.sender, Mechanism):
294	# owner_mech = self.sender
295	# else:
296	# raise AutoAssociativeError("The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently"
297	# " the sender is {}".
298	# format(self.__class__.__name__, self.name, self.sender))
299	# if AUTO in owner_mech._parameter_ports and HETERO in owner_mech._parameter_ports:
300	# owner_mech._parameter_ports[AUTO].update(context=context, params=runtime_params, time_scale=time_scale)
301	# owner_mech._parameter_ports[HETERO].update(context=context, params=runtime_params, time_scale=time_scale)
302	#
303
304	# END OF COMMENTED OUT BY KAM 1/9/2018
305
306	# NOTE 7/25/17 CW: Originally, this override was written because if the user set the 'auto' parameter on the
307	# recurrent mechanism, the ParameterPort wouldn't update until after the mechanism executed: since the system
308	# first runs the projection, then runs the mechanism, the projection initially uses the 'old' value. However,
309	# this is commented out because this may in fact be the desired behavior.
310	# Two possible solutions: allow control to be done on projections, or build a more general way to allow
311	# projections to read parameters from mechanisms.
312	# def _update_parameter_ports(self, runtime_params=None, context=None):
313	# """Update this projection's owner mechanism's `auto` and `hetero` parameter ports as well! The owner mechanism
314	# should be a RecurrentTransferMechanism, which DOES NOT update its own `auto` and `hetero` parameter ports during
315	# its _update_parameter_ports function (so that the ParameterPort is not redundantly updated).
316	# Thus, if you want to have an AutoAssociativeProjection on a mechanism that's not a RecurrentTransferMechanism,
317	# your mechanism must similarly exclude `auto` and `hetero` from updating.
318	# """
319	# super()._update_parameter_ports(runtime_params, context)
320	#
321	# if isinstance(self.sender, OutputPort):
322	# owner_mech = self.sender.owner
323	# elif isinstance(self.sender, Mechanism):
324	# owner_mech = self.sender
325	# else:
326	# raise AutoAssociativeError("The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently the"
327	# " sender is {}".
328	# format(self.__class__.__name__, self.name, self.sender))
329	#
330	# if AUTO in owner_mech._parameter_ports and HETERO in owner_mech._parameter_ports:
331	# owner_mech._parameter_ports[AUTO].update(context=context, params=runtime_params)
332	# owner_mech._parameter_ports[HETERO].update(context=context, params=runtime_params)
333	# else:
334	# raise AutoAssociativeError("Auto or Hetero ParameterPort not found in {0} \"{1}\"; here are names of the "
335	# "current ParameterPorts for {1}: {2}".format(owner_mech.__class__.__name__,
336	# owner_mech.name, owner_mech._parameter_ports.key_values))
337
338	@property	1✔
339	def owner_mech(self):	1✔
340	if isinstance(self.sender, OutputPort):	1!
341	return self.sender.owner	1✔
342	elif isinstance(self.sender, Mechanism):	×
343	return self.sender	×
344	else:
345	raise AutoAssociativeError(
346	"The sender of the {} \'{}\' must be a Mechanism or OutputPort: currently the sender is {}".format(
347	self.__class__.__name__, self.name, self.sender
348	)
349	)
350
351	@property	1✔
352	def matrix(self):	1✔
353	owner_mech = self.owner_mech	1✔
354	if hasattr(owner_mech, "matrix"):	1!
355	return owner_mech.matrix	1✔
356	return super(AutoAssociativeProjection, self.__class__).matrix.fget(self)	×
357
358	@matrix.setter	1✔
359	def matrix(self, setting):	1✔
360	owner_mech = self.owner_mech	1✔
361	if hasattr(owner_mech, "matrix"):	1!
362	owner_mech.matrix = setting	1✔
363	else:
364	super(AutoAssociativeProjection, self.__class__).matrix.fset(self, setting)	×
365
366
367	# a helper function that takes a specification of `hetero` and returns a hollow matrix with the right values
368	def get_hetero_matrix(raw_hetero, size):	1✔
369	if is_numeric_scalar(raw_hetero):	1✔
370	return get_matrix(HOLLOW_MATRIX, size, size) * raw_hetero	1✔
371	elif ((isinstance(raw_hetero, np.ndarray) and raw_hetero.ndim == 1) or	1!
372	(isinstance(raw_hetero, list) and np.array(raw_hetero).ndim == 1)):
373	if len(raw_hetero) != 1:	×
374	return None	×
375	return get_matrix(HOLLOW_MATRIX, size, size) * raw_hetero[0]	×
376	elif (isinstance(raw_hetero, np.matrix) or	1✔
377	(isinstance(raw_hetero, np.ndarray) and raw_hetero.ndim == 2) or
378	(isinstance(raw_hetero, list) and np.array(raw_hetero).ndim == 2)):
379	np.fill_diagonal(raw_hetero, 0)	1✔
380	return np.array(raw_hetero)	1✔
381	else:
382	return None	1✔
383
384
385	# similar to get_hetero_matrix() above
386	def get_auto_matrix(raw_auto, size):	1✔
387	if is_numeric_scalar(raw_auto):	1✔
388	return np.diag(np.full(size, raw_auto, dtype=float))	1✔
389	elif ((isinstance(raw_auto, np.ndarray) and raw_auto.ndim == 1) or	1✔
390	(isinstance(raw_auto, list) and np.array(raw_auto).ndim == 1)):
391	if len(raw_auto) == 1:	1✔
392	return np.diag(np.full(size, raw_auto[0], dtype=float))	1✔
393	else:
394	if len(raw_auto) != size:	1!
UNCOV 395	return None	×
396	return np.diag(raw_auto)	1✔
397	elif (isinstance(raw_auto, np.matrix) or	1!
398	(isinstance(raw_auto, np.ndarray) and raw_auto.ndim == 2) or
399	(isinstance(raw_auto, list) and np.array(raw_auto).ndim == 2)):
400	# we COULD add a validation here to ensure raw_auto is diagonal, but it would slow stuff down.
401	return np.array(raw_auto)	×
402	else:
403	return None	1✔

PrincetonUniversity / PsyNeuLink / 11972746306

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous