def test_forwardPropagation(self):
for example, predicted in zip(examples, predicteds):
ex = toVector(example)
pred = toVector(predicted)
assert_array_almost_equal(self.network.forwardPropagation(ex),
pred,
decimal=5)
def test_J(self):
for i, (output, predicted) in enumerate(zip(outputs, predicteds)):
out = toVector(output)
pred = toVector(predicted)
self.network.updateCost(out, pred)
self.assertAlmostEqual(self.network.J, J[i], places=3)
self.network.reset()
def test_cost_computation(self):
for output, predicted in zip(outputs, predicteds):
out = toVector(output)
pred = toVector(predicted)
self.network.updateCost(out, pred)
self.assertAlmostEqual(self.network.computeCost(len(outputs)),
Jtotal,
places=4)
def test_upgradeGrads(self):
for i, (example, output) in enumerate(zip(examples, outputs)):
ex = toVector(example)
out = toVector(output)
pred = self.network.forwardPropagation(ex)
self.network.backPropagation(pred - out)
self.network.updateGrads(len(examples))
for i, D in enumerate(self.network.D):
assert_array_almost_equal(D, np.array(gradsTotal[i]), decimal=5)
def test_backPropagation(self):
for i, (example, output) in enumerate(zip(examples, outputs)):
ex = toVector(example)
out = toVector(output)
pred = self.network.forwardPropagation(ex)
self.network.backPropagation(pred - out)
for j, D in enumerate(self.network.D):
assert_array_almost_equal(D, np.array(grads[i][j]), decimal=5)
self.network.reset()
def test_trainTurn(self):
datapoints = [(toVector(input), toVector(output))
for input, output in zip(examples, outputs)]
self.network.trainTurn(datapoints)
for i, D in enumerate(self.network.D):
assert_array_almost_equal(D, np.array(gradsTotal[i]), decimal=5)
err = self.network.trainTurn(datapoints, updateCost=True)
self.assertAlmostEqual(self.network.computeCost(len(examples)),
Jtotal,
places=2)
def test_toVector(self):
attr1 = [1.0, 0.56, 0]
assert_array_almost_equal(toVector(attr1),
np.array([[1.0], [0.56], [0]]),
decimal=5)