def test_base_creation(self):
neurons = list()
for i in range(0, 10):
neurons.append(snn.Neuron("linear", _np.random.rand(43)))
l = snn.Layer()
l.add_neurons(neurons)
# Need this to make layer create matrix.
for neuron in l:
pass
self.assertEqual(l.out_len(), 10)
self.assertEqual(l.in_len(), 43)
def test_layer_fails_on_bad_input_003(self):
neurons = list()
for i in range(0, 148):
neurons.append(snn.Neuron("linear", _np.random.rand(44)))
l = snn.Layer()
l.add_neurons(neurons)
for n in l:
pass
input = _np.array([])
self.assertRaises(ValueError, lambda: l.input(input))
def test_iterable(self):
neuron_count = 10
neurons = list()
for i in range(0, neuron_count):
neurons.append(snn.Neuron("linear", _np.random.rand(43)))
l = snn.Layer()
l.add_neurons(neurons)
count = 0
for neuron in l:
count += 1
self.assertEqual(count, neuron_count)
def test_func_weights_give_right_output(self):
l = snn.Layer()
for i in range(0, 5):
l.add_neurons(snn.Neuron("npower", func_weights_count=1))
input = _np.array([1, 1, 1])
good_output = _np.array([3.0, 9.0, 27.0, 81.0, 243.0]).reshape(5, 1)
for neuron in l:
neuron.set_input_weights(_np.array([1.0, 1.0, 1.0]))
neuron.set_func_weights(_np.array(0.0))
func_weights = _np.array([1, 2, 3, 4, 5])
l.set_func_weights(func_weights)
real_output = l.input(input.reshape(3, 1))
self.assertEquals(real_output.size, 5)
_np_tests.assert_array_equal(real_output, good_output)
def test_layer_get_weights(self):
weights = _np.array([5.0, 5.0, 5.0])
func_weights = _np.array([34.0])
neurons = list()
for i in range(0, 150):
neurons.append(snn.Neuron("linear", weights.copy(), func_weights.copy()))
l = snn.Layer()
l.add_neurons(neurons)
for n in l:
pass
actual_weights = l.get_weights("input")
actual_func_weights = l.get_weights("func")
actual_all_weights = l.get_weights("all")
_np_tests.assert_equal(actual_weights, 5.0)
_np_tests.assert_equal(actual_func_weights, 34.0)
self.assertEqual(actual_all_weights.size, actual_weights.size + actual_func_weights.size)
def load_from_json(json_str):
net_dump = _json.loads(json_str)
input_count = net_dump["input_count"]
error_name = net_dump["error_name"]
net = PerceptronNumpy(input_count, error_name)
for layer_dump in net_dump["layers"]:
layer = _snn.Layer()
for neuron_dump in layer_dump:
if "weights" in neuron_dump:
weights = _np.array(neuron_dump["weights"])
else:
weights = None
if "func_weights" in neuron_dump:
func_weights = _np.array(neuron_dump["func_weights"])
else:
func_weights = None
neuron = _snn.Neuron(neuron_dump["func_name"], weights,
func_weights, neuron_dump["f_len"])
layer.add_neurons(neuron)
net.add_layer(layer)
return net
def test_layer_set_weights(self):
weights = _np.array([5.0, 5.0, 5.0])
func_weights = _np.array([34.0])
neurons = list()
for i in range(0, 150):
neurons.append(
snn.Neuron("linear", weights.copy(), func_weights.copy()))
l = snn.Layer()
l.add_neurons(neurons)
for n in l:
pass
weights = l.get_weights("input")
weights.fill(0)
func_weights = l.get_weights("func")
func_weights.fill(-.4)
l.set_weights(weights.copy())
_np_tests.assert_equal(l.get_weights("input"), 0)
l.set_func_weights(func_weights)
_np_tests.assert_equal(l.get_weights("func"), -0.4)
def test_layer_returns_right_output(self):
neuron_count = 45
input_len = 56
neurons = list()
for i in range(0, neuron_count):
weights = _np.zeros(input_len, dtype=_np.float64)
weights.fill(1.0)
neurons.append(snn.Neuron("linear", weights))
l = snn.Layer()
l.add_neurons(neurons)
for n in l:
pass
input = _np.zeros(input_len, dtype=_np.float64).reshape(input_len, 1)
input.fill(2.0)
expected_output = 2.0 * input_len
expected_array = _np.zeros(neuron_count, dtype=_np.float64)
expected_array.fill(expected_output)
real_array = l.input(input).A1
_np_tests.assert_allclose(real_array, expected_array)
def test_add_neurons_raises(self):
l = snn.Layer()
self.assertRaises(ValueError, lambda: l.add_neurons("test"))