def test_neural_tree_network(self):
net = NeuralTreeNet(3, empty=False)
X = numpy.random.randn(2, 3)
got = net.predict(X)
exp = X.sum(axis=1)
self.assertEqual(exp.reshape((-1, 1)), got[:, -1:])
rep = repr(net)
self.assertEqual(rep, 'NeuralTreeNet(3)')
net.clear()
self.assertEqual(len(net), 0)
def test_convert(self):
X = numpy.arange(8).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 10).astype(numpy.int64)
y2 = y.copy()
y2[0] = 2
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y2)
self.assertRaise(
lambda: NeuralTreeNet.create_from_tree(tree), RuntimeError)
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10)
self.assertNotEmpty(root)
exp = tree.predict_proba(X)
got = root.predict(X)
self.assertEqual(exp.shape[0], got.shape[0])
self.assertEqualArray(exp, got[:, -2:])
def test_neural_net_gradient(self):
X = numpy.arange(8).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 10).astype(numpy.int64)
ny = label_class_to_softmax_output(y)
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10)
_, out, err = self.capture(lambda: root.fit(X, ny, verbose=True))
self.assertIn("loss:", out)
self.assertEmpty(err)
def test_neural_tree_network_copy(self):
net = NeuralTreeNet(3, empty=False)
net.append(NeuralTreeNode(1, activation='identity'),
inputs=[3])
net2 = net.copy()
X = numpy.random.randn(2, 3)
self.assertEqualArray(net.predict(X), net2.predict(X))
def test_convert_compact(self):
X = numpy.arange(8).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 10).astype(numpy.int64)
y2 = y.copy()
y2[0] = 2
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y2)
self.assertRaise(
lambda: NeuralTreeNet.create_from_tree(tree, arch="k"), ValueError)
self.assertRaise(
lambda: NeuralTreeNet.create_from_tree(tree, arch="compact"),
RuntimeError)
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10, arch='compact')
self.assertNotEmpty(root)
exp = tree.predict_proba(X)
got = root.predict(X)
self.assertEqual(exp.shape[0], got.shape[0])
self.assertEqualArray(exp + 1e-8, got[:, -2:] + 1e-8)
dot = root.to_dot()
self.assertIn("s3a4:f4 -> s5a6:f6", dot)
def test_neural_net_gradient_regression_2(self):
X = numpy.abs(numpy.random.randn(10, 2))
w1 = numpy.array([-0.5, 0.8, -0.6])
noise = numpy.random.randn(X.shape[0]) / 10
noise[0] = 0
noise[1] = 0.07
X[1, 0] = 0.7
X[1, 1] = -0.5
y = w1[0] + X[:, 0] * w1[1] + X[:, 1] * w1[2] + noise
for act in [
'relu', 'sigmoid', 'identity', 'leakyrelu', 'sigmoid4', 'expit'
]:
with self.subTest(act=act):
neu = NeuralTreeNode(w1[1:], bias=w1[0], activation=act)
loss1 = neu.loss(X, y)
pred1 = neu.predict(X)
if act == 'relu':
self.assertEqualArray(pred1[1:2], numpy.array([0.36]))
pred11 = neu.predict(X)
self.assertEqualArray(pred11[1:2], numpy.array([0.36]))
net = NeuralTreeNet(X.shape[1], empty=True)
net.append(neu, numpy.arange(0, 2))
ide = NeuralTreeNode(numpy.array([1], dtype=X.dtype),
bias=numpy.array([0], dtype=X.dtype),
activation='identity')
net.append(ide, numpy.arange(2, 3))
pred2 = net.predict(X)
loss2 = net.loss(X, y)
self.assertEqualArray(pred1, pred2[:, -1])
self.assertEqualArray(pred2[:, -2], pred2[:, -1])
self.assertEqualArray(pred2[:, 2], pred2[:, 3])
self.assertEqualArray(loss1, loss2)
for p in range(0, 5):
grad1 = neu.gradient(X[p], y[p])
grad2 = net.gradient(X[p], y[p])
self.assertEqualArray(grad1, grad2[:3])
def test_training_weights(self):
X = numpy.arange(8).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 10).astype(numpy.int64)
y2 = y.copy()
y2[0] = 2
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10)
v1 = root.predict(X[:1])
w = root.training_weights
self.assertEqual(w.shape, (11, ))
delta = numpy.arange(11) + 0.5
root.update_training_weights(delta)
v2 = root.predict(X[:1])
self.assertNotEqualArray(v1, v2)
def test_neural_tree_network_training_weights(self):
net = NeuralTreeNet(3, empty=False)
net.append(NeuralTreeNode(1, activation='identity'), inputs=[3])
w = net.training_weights
self.assertEqual(w.shape, (6, ))
self.assertEqual(w[0], 0)
self.assertEqualArray(w[1:4], [1, 1, 1])
delta = numpy.arange(6) - 0.5
net.update_training_weights(delta)
w2 = net.training_weights
self.assertEqualArray(w2, w + delta)
def test_neural_net_gradient_fit(self):
X = numpy.arange(16).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 15).astype(numpy.int64)
ny = label_class_to_softmax_output(y)
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10)
loss1 = root.loss(X, ny).sum()
self.assertGreater(loss1, -1e-5)
self.assertLess(loss1, 1.)
_, out, err = self.capture(
lambda: root.fit(X, ny, verbose=True, max_iter=20))
self.assertEmpty(err)
self.assertNotEmpty(out)
loss2 = root.loss(X, ny).sum()
self.assertLess(loss2, loss1 + 1)
def test_neural_tree_network_append(self):
net = NeuralTreeNet(3, empty=False)
self.assertRaise(lambda: net.append(
NeuralTreeNode(2, activation='identity'), inputs=[3]))
net.append(NeuralTreeNode(1, activation='identity'), inputs=[3])
self.assertEqual(net.size_, 5)
last_node = net.nodes[-1]
X = numpy.random.randn(2, 3)
got = net.predict(X)
exp = X.sum(axis=1) * last_node.input_weights[0] + last_node.bias
self.assertEqual(exp.reshape((-1, 1)), got[:, -1:])
rep = repr(net)
self.assertEqual(rep, 'NeuralTreeNet(3)')
def test_convert_compact_fit(self):
X = numpy.arange(8).astype(numpy.float64).reshape((-1, 2))
y = ((X[:, 0] + X[:, 1] * 2) > 10).astype(numpy.int64)
y2 = y.copy()
y2[0] = 2
tree = DecisionTreeClassifier(max_depth=2)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree, 10, arch='compact')
self.assertNotEmpty(root)
exp = tree.predict_proba(X)
got = root.predict(X)
self.assertEqual(exp.shape[0], got.shape[0])
self.assertEqualArray(exp + 1e-8, got[:, -2:] + 1e-8)
ny = label_class_to_softmax_output(y)
loss1 = root.loss(X, ny).sum()
_, out, err = self.capture(
lambda: root.fit(X, ny, verbose=True, max_iter=20))
self.assertEmpty(err)
self.assertNotEmpty(out)
loss2 = root.loss(X, ny).sum()
self.assertLess(loss2, loss1 + 1)
def test_dot(self):
data = load_iris()
X, y = data.data, data.target
y = y % 2
tree = DecisionTreeClassifier(max_depth=3, random_state=11)
tree.fit(X, y)
root = NeuralTreeNet.create_from_tree(tree)
dot = export_graphviz(tree)
self.assertIn("digraph", dot)
dot2 = root.to_dot()
self.assertIn("digraph", dot2)
x = X[:1].copy()
x[0, 3] = 1.
dot2 = root.to_dot(X=x.ravel())
self.assertIn("digraph", dot2)
exp = tree.predict_proba(X)[:, -1]
got = root.predict(X)[:, -1]
mat = numpy.empty((exp.shape[0], 2), dtype=exp.dtype)
mat[:, 0] = exp
mat[:, 1] = got
c = numpy.corrcoef(mat.T)
self.assertGreater(c[0, 1], 0.5)
def test_neural_net_gradient_regression(self):
X = numpy.abs(numpy.random.randn(10, 2))
w1 = numpy.array([-0.5, 0.8, -0.6])
noise = numpy.random.randn(X.shape[0]) / 10
noise[0] = 0
noise[1] = 0.07
X[1, 0] = 0.7
X[1, 1] = -0.5
y = w1[0] + X[:, 0] * w1[1] + X[:, 1] * w1[2] + noise
for act in ['identity', 'relu', 'leakyrelu',
'sigmoid', 'sigmoid4', 'expit']:
with self.subTest(act=act):
neu = NeuralTreeNode(w1[1:], bias=w1[0], activation=act)
loss1 = neu.loss(X, y)
grad1 = neu.gradient(X[0], y[0])
net = NeuralTreeNet(X.shape[1], empty=True)
net.append(neu, numpy.arange(0, 2))
loss2 = net.loss(X, y)
grad2 = net.gradient(X[0], y[0])
self.assertEqualArray(loss1, loss2)
self.assertEqualArray(grad1, grad2)
