def test_learn_synapse_connections(self):
active = np.array([[1, 1], [1, 1]])
input_vector = np.array([[1, 0], [0, 0]])
p_inc = 0.1
p_dec = 0.1
part_deque = partial(RingBuffer, input_array=np.zeros(2), copy=True)
activity = defaultdict(part_deque)
overlap_sum = defaultdict(part_deque)
min_activity = np.array([[1, 0], [0, 0]])
boost = np.array([[1, 1], [1, 1]])
b_inc = 1
p_mult = 1
columns, synapse_modified =\
BSP.learn_synapse_connections(self.columns, active,
input_vector, p_inc, p_dec,
activity, overlap_sum,
min_activity, boost, b_inc,
p_mult)
self.assertIsInstance(columns, np.ndarray)
self.assertEqual(columns.shape, (2, 2, 2, 2))
self.assertListEqual(columns.tolist(), [
[
[[0.1, 0.], [0., 0.]],
[[0.1, 0.], [0., 0.]]
],
[
[[0.1, 0.], [0., 0.]],
[[0.1, 0.], [0., 0.]]
]
])
self.assertIsInstance(synapse_modified, types.BooleanType)
self.assertFalse(synapse_modified)
def test_spatial_pooler(self,):
images = np.zeros(shape=(10, 2, 2))
shape = (2, 2, 2, 2)
p_connect = 0.15
connect_threshold = 0.2
p_inc = 0.02
p_dec = 0.02
b_inc = 0.005
p_mult = 0.01
min_activity_threshold = 0.01
min_overlap = 3,
desired_activity_mult = 0.05
max_iterations = 1000
cols = BSP.spatial_pooler(images, shape, p_connect,
connect_threshold, p_inc,
p_dec, b_inc, p_mult,
min_activity_threshold, min_overlap,
desired_activity_mult,
max_iterations)
self.assertIsInstance(cols, np.ndarray)
self.assertEqual(cols.shape, shape)
expr = ne.evaluate('0 <= cols')
self.assertTrue((expr | np.isnan(cols)).all())
expr = ne.evaluate('cols <= 1')
self.assertTrue((expr | np.isnan(cols)).all())
def test_calculate_overlap(self):
input_vector = np.array([[1, 0], [0, 0]])
self.columns[0, 0, 0, 0] = 1
min_overlap = 1
connect_threshold = 0.1
boost = np.array([[1, 1], [1, 1]])
part_deque = partial(RingBuffer, input_array=np.zeros(2), copy=True)
overlap_sum = defaultdict(part_deque)
overlap, overlap_sum =\
BSP.calculate_overlap(input_vector, self.columns, min_overlap,
connect_threshold, boost, overlap_sum)
self.assertIsInstance(overlap, np.ndarray)
self.assertEqual(overlap.shape, (2, 2))
self.assertIsInstance(overlap_sum, defaultdict)
self.assertListEqual(overlap.tolist(), [[1., 0.], [0., 0.]])
self.assertIsInstance(overlap_sum[0, 0], RingBuffer)
self.assertListEqual(list(overlap_sum[0, 0]), [0, 1])
