def epnet_run():
epnet = EPNet(10, 5, 5, 1)
genp5 = ParityNGenerator(5)
_, res, vec = map(np.array, zip(*genp5.all()))
top, i = epnet.run(vec, res, num_hid=2, lr=0.5, temperature=1.0)
print(top)
print(i)
def test_evaluate_given7(self):
weights_out = '''-42.8 -32.4 0 -5.6 -23.6 -33.6 -33.6 41.6 0 0 0
-75.3 -32.0 43.2 -41.1 -34.5 -34.8 -34.8 39.8 -58.9 0 0
-85.0 -28.1 28.6 -28.0 -28.0 -28.2 -28.2 29.3 -47.6 -41.3 0
59.9 12.9 -13.5 13.0 13.0 13.0 13.0 -13.4 0 0 81.8
'''
weights = np.fromstring(weights_out, sep=' ').reshape(4, -1)
ann = ForwardArtificialNeuralNectwork(7, 7, 1)
ann.construct(weights)
genp7 = ParityNGenerator(7)
for _, res, vec in genp7.all():
result = ann.evaluate(vec)
self.assertEqual(result>0.5, not res) # use not res since the condition is negating in paper
def test_evaluate_given8(self):
weights_out = '''-12.4 25.2 27.7 -29.4 -28.9 -29.7 -25.4 -28.5 27.8 0 0 0
-40.4 19.6 18.9 -18.1 -19.1 -18.5 -17.3 -18.8 20.4 -67.6 0 0
-48.1 16.0 16.1 -15.9 -16.3 -15.8 -15.9 -15.8 16.7 -55.0 -26.7 0
45.7 -10.0 -11.0 10.0 9.9 9.4 10.0 9.6 -11.4 6.8 2.3 76.3
'''
weights = np.fromstring(weights_out, sep=' ').reshape(4, -1)
ann = ForwardArtificialNeuralNectwork(8, 8, 1)
ann.construct(weights)
genp8 = ParityNGenerator(8)
_, res, vec = zip(*genp8.all())
result = ann.evaluate(np.array(vec))
for yhat, y in zip(result, res):
self.assertEqual(yhat>0.5, not y) # the same as the case given7
def test_train_given6(self):
# mean square error function
mse = lambda y1,y2: ((y1-y2)**2/2).sum()
for _ in range(10):
ann = ForwardArtificialNeuralNectwork(6, 6, 1)
ann.initialize(3, 1.0, 0, 1)
genp6 = ParityNGenerator(6)
_, res, vec = map(np.array, zip(*genp6.all()))
before = ann.evaluate(vec)
be_mse = mse(before, res)
ann.train(vec, res, 0.1, 1000)
after = ann.evaluate(vec)
af_mse = mse(after, res)
self.assertLessEqual(af_mse, be_mse)
def test_call(self):
pn = ParityNGenerator(4)
vec, exp = pn(11)
self.assertTupleEqual(tuple(vec), (True, False, True, True))
self.assertEqual(exp, False)
with self.assertRaises(ParityNGenerator.ParityNException):
pn(16)
with self.assertRaises(ParityNGenerator.ParityNException):
pn(4, 1, a=3)
def ann_run():
ann = ForwardArtificialNeuralNectwork(5, 5, 1)
ann.initialize(2, 1.0, 0.0, 1.0)
genp5 = ParityNGenerator(5)
_, res, vec = map(np.array, zip(*genp5.all()))
ann.train(vec, res, lr=0.3, epoch=1000)
result = (ann.evaluate(vec) > 0.5)
for y, yhat in zip(res, result):
if y != yhat:
print(y, yhat)
print()
ann.simul_anneal(vec, res, temperature=1.0, steps=1000)
result = (ann.evaluate(vec) > 0.5)
for y, yhat in zip(res, result):
if y != yhat:
print(y, yhat)
def test_all(self):
it = ParityNGenerator(2).all()
_, out, vec = next(it)
self.assertEqual(out, np.array([True]))
self.assertTupleEqual(tuple(vec), (False, False))
_, out, vec = next(it)
self.assertEqual(out[0], False)
self.assertTupleEqual(tuple(vec), (False, True))
_, out, vec = next(it)
self.assertEqual(out[0], False)
self.assertTupleEqual(tuple(vec), (True, False))
_, out, vec = next(it)
self.assertEqual(out[0], True)
self.assertTupleEqual(tuple(vec), (True, True))
def test_constructor(self):
pn = ParityNGenerator(6)
self.assertEqual(pn.bound, 64)
with self.assertRaises(ParityNGenerator.ParityNException):
ParityNGenerator(-1)