PrincetonUniversity / PsyNeuLink / 6757715192

84%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

Refactor/learning pathways using ports (#2827)

* • composition.py
   - refactor learning pathways to be specific to input_port and output_ports
     (to accomodate parallel pathways through Mechanisms with TransferFunction:
   - add _get_ports_for_input_output_sources()
   - _create_terminal_backprop_learning_components()
     _create_learning_related_mechanisms()
     _create_learning_related_projections():
       all take input_source_output_port and output_source_input_port as args
  - _add_error_projection_to_dependent_learning_mechs,
    _create_non_terminal_backprop_learning_components:
      - separate error_sources for parallel InputPort/OutputPort "streams" in a single Mechanism that use TransferFunctions
      - if mechanism uses TransferFunction, only consider dependent_learning_projections
         for projections to input_port with same index as output_port of error_source
  - _create_backpropagation_learning_pathway():
      - determine dependent_learning_mechanisms locally
      - raise error for terminal node with > 1 output_ports
  - _instantiate_input_dict():
      refactor to format user-specified inputs

  - _create_backpropagation_learning_pathway():


• autodiffcomposition.py, pytorchcomponents.py:
  standardize on value of PytorchMechanismWrapper as 2d
  to accomodate >1 output_ports

• learningmechanism.py
    remove dependent_learning_mechanisms property

• processingmechanism.py:
  - _instantiate_output_ports():
     if no OutputPorts are specified, and variable has more than one item, assign one OutputPort to each
  - docstring description of configuration


• test_composition.py
  add test_input_shape_errors

• test_autodiffcomposition.py
  - add test_two_output_ports_on_OUTPUT_Node

• test_learning.py
  - TestAutodiffMultipleOutput_port:
    - add test_parallel_inputs_to_output_ports_converge_internal
    - add test_single_input_to_multiple_output_ports_converge_internal
    - ... (continued)

Run Details

13302 of 16412 branches covered (0.0%)

Branch coverage included in aggregate %.

210 of 218 new or added lines in 7 files covered. (96.33%)

3 existing lines in 2 files now uncovered.

31299 of 36135 relevant lines covered (86.62%)

0.87 hits per line