def test_backprop():
"""
Test for no hidden layers, with lr=0.005 and y=10
"""
Net = neuralNet([1, 1], batchSize=1)
Net.applyNet([5])
Net.w[0] = 2.0
Net.b[0] = 3.0
y_act = 10
lr = 0.005
y_pred = (2.0 * 5.0) + 3.0
y_pred = y_pred * (y_pred > 0)
cost = 0.5 * (y_act - y_pred)**2
dCdy = (-y_pred)
dydz = (y_pred > 0)
dzdw = 5
dzdb = 1
dCdw = dCdy * dydz * dzdw
dCdb = dCdy * dydz * dzdb
w0 = 2.0 - (lr * dCdw)
b0 = 3.0 - (lr * dCdb)
Net.trainNet([5], y_act, lr=0.005)
np.testing.assert_approx_equal(Net.w[0], 1.925)
np.testing.assert_approx_equal(Net.b[0], 2.985)
def test_feedforward1():
"""
test for a single layer, no hidden layers
"""
Net = neuralNet([1, 1])
Net.w[0] = 2.0
Net.b[0] = 3.0
act_val = (2.0 * 3.0) + 3.0
act_val = act_val * (act_val > 0)
assert Net.applyNet([3]) == act_val
def test_feedforward4():
"""
test for a single layer, no hidden layers
testinf for negative values
"""
Net = neuralNet([1, 1])
Net.w[0] = 2.0
Net.b[0] = 3.0
act_val = (2.0 * -5.0) + 3.0
act_val = act_val * (act_val > 0)
assert Net.applyNet([-5]) == act_val
def test_feedforward3():
"""
test for 0 hidden layers, two input neurons
"""
Net = neuralNet([2, 1])
Net.w[0][0, 0] = 2.0
Net.b[0][0] = 3.0
Net.w[0][1, 0] = 4.0
act_val = (6.0 * 2.0) + (7.0 * 4.0) + 3.0
act_val = act_val * (act_val > 0)
assert Net.applyNet([6, 7]) == act_val
def test_feedforward2():
"""
test for 1 hidden layers, one neuron each
"""
Net = neuralNet([1, 1, 1])
Net.w[0] = 2.0
Net.b[0] = 3.0
Net.w[1] = 4.0
Net.b[1] = 5.0
act_val = 4.0 * (2.0 * 6 + 3.0) + 5.0
act_val = act_val * (act_val > 0)
assert Net.applyNet([6]) == act_val
def Net1():
Net = neuralNet([5, 4, 3, 2])
return Net