def test_recurrent_mech_with_learning(self):
R = RecurrentTransferMechanism(size=4,
function=Linear,
matrix=np.full((4, 4), 0.1),
enable_learning=True)
# Test that all of these are the same:
np.testing.assert_allclose(
R.recurrent_projection.mod_matrix,
[[0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1],
[0.1, 0.1, 0.1, 0.1]])
np.testing.assert_allclose(R.recurrent_projection.matrix, R.matrix)
np.testing.assert_allclose(R.input_state.path_afferents[0].matrix,
R.matrix)
# Test that activity is properly computed prior to learning
p = Process(pathway=[R])
R.learning_enabled = False
p.execute([1, 1, 0, 0])
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(R.value, [[1.2, 1.2, 0.2, 0.2]])
# Test that activity and weight changes are properly computed with learning
R.learning_enabled = True
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(R.value, [[1.28, 1.28, 0.28, 0.28]])
np.testing.assert_allclose(R.recurrent_projection.mod_matrix, [[
0.1, 0.18192000000000003, 0.11792000000000001, 0.11792000000000001
], [
0.18192000000000003, 0.1, 0.11792000000000001, 0.11792000000000001
], [
0.11792000000000001, 0.11792000000000001, 0.1, 0.10392000000000001
], [
0.11792000000000001, 0.11792000000000001, 0.10392000000000001, 0.1
]])
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(
R.value, [[1.4268928, 1.4268928, 0.3589728, 0.3589728]])
np.testing.assert_allclose(R.recurrent_projection.mod_matrix,
[[0.1, 0.28372115, 0.14353079, 0.14353079],
[0.28372115, 0.1, 0.14353079, 0.14353079],
[0.14353079, 0.14353079, 0.1, 0.11036307],
[0.14353079, 0.14353079, 0.11036307, 0.1]])
def test_recurrent_mech_with_learning(self):
R = RecurrentTransferMechanism(size=4,
function=Linear,
matrix=np.full((4, 4), 0.1),
enable_learning=True)
# Test that all of these are the same:
np.testing.assert_allclose(
R.matrix, [[0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1],
[0.1, 0.1, 0.1, 0.1], [0.1, 0.1, 0.1, 0.1]])
np.testing.assert_allclose(R.recurrent_projection.matrix, R.matrix)
np.testing.assert_allclose(R.input_state.path_afferents[0].matrix,
R.matrix)
# Test that activity is properly computed prior to learning
p = Process(pathway=[R])
R.learning_enabled = False
p.execute([1, 1, 0, 0])
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(R.value, [[1.2, 1.2, 0.2, 0.2]])
# Test that activity and weight changes are properly computed with learning
R.learning_enabled = True
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(R.value, [[1.28, 1.28, 0.28, 0.28]])
np.testing.assert_allclose(
R.matrix, [[
0.18192000000000003, 0.18192000000000003, 0.11792000000000001,
0.11792000000000001
],
[
0.18192000000000003, 0.18192000000000003,
0.11792000000000001, 0.11792000000000001
],
[
0.11792000000000001, 0.11792000000000001,
0.10392000000000001, 0.10392000000000001
],
[
0.11792000000000001, 0.11792000000000001,
0.10392000000000001, 0.10392000000000001
]])
p.execute([1, 1, 0, 0])
np.testing.assert_allclose(
R.value, [[1.5317504, 1.5317504, 0.3600704, 0.3600704]])
np.testing.assert_allclose(
R.matrix, [[
0.299232964395008, 0.299232964395008, 0.14549689896140802,
0.14549689896140802
],
[
0.299232964395008, 0.299232964395008,
0.14549689896140802, 0.14549689896140802
],
[
0.14549689896140802, 0.14549689896140802,
0.11040253464780801, 0.11040253464780801
],
[
0.14549689896140802, 0.14549689896140802,
0.11040253464780801, 0.11040253464780801
]])