def test_shuffle():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=1,
shuffle_init=True,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert round(acc, 2) == 0.99, "Acc: %s" % acc
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=1,
shuffle_init=True,
shuffle_epoch=True,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert round(acc, 2) == 0.99, "Acc: %s" % acc
def test_shuffle():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=1,
shuffle_init=True,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert round(acc, 2) == 0.99, "Acc: %s" % acc
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=1,
shuffle_init=True,
shuffle_epoch=True,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert round(acc, 2) == 0.99, "Acc: %s" % acc
def test_gradient_descent():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=10,
eta=0.01,
minibatches=1,
shuffle=True,
random_state=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
#print(acc)
assert(round(acc, 2) == 0.87)
def test_binary():
X0 = X_std[0:100] # class 0 and class 1
y0 = y[0:100] # class 0 and class 1
nn = NeuralNetMLP(n_output=2,
n_features=X0.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=10,
shuffle_init=True,
shuffle_epoch=True,
random_seed=1)
nn.fit(X0, y0)
y_pred = nn.predict(X0)
acc = np.sum(y0 == y_pred, axis=0) / float(X0.shape[0])
assert round(acc, 2) == 1.0, "Acc: %s" % acc
def test_gradient_descent():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
random_weights=[-1.0, 1.0],
minibatches=1,
shuffle_init=False,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert(round(acc, 2) == 0.97)
def test_minibatch():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=15,
alpha=2.0,
eta=0.05,
minibatches=10,
shuffle_init=True,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
print(acc)
assert(round(acc, 2) == 0.97)
def test_binary():
X0 = X_std[0:100] # class 0 and class 1
y0 = y[0:100] # class 0 and class 1
nn = NeuralNetMLP(n_output=2,
n_features=X0.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
minibatches=10,
shuffle_init=True,
shuffle_epoch=True,
random_seed=1)
nn.fit(X0, y0)
y_pred = nn.predict(X0)
acc = np.sum(y0 == y_pred, axis=0) / float(X0.shape[0])
assert round(acc, 2) == 1.0, "Acc: %s" % acc
def test_minibatch():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=15,
alpha=2.0,
eta=0.05,
minibatches=10,
shuffle_init=True,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
print(acc)
assert (round(acc, 2) == 0.97)
def test_gradient_descent():
nn = NeuralNetMLP(n_output=3,
n_features=X.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=100,
eta=0.1,
random_weights=[-1.0, 1.0],
minibatches=1,
shuffle_init=False,
shuffle_epoch=False,
random_seed=1)
nn.fit(X_std, y)
y_pred = nn.predict(X_std)
acc = np.sum(y == y_pred, axis=0) / float(X_std.shape[0])
assert (round(acc, 2) == 0.97)
def test_binary():
X0 = X_std[0:100] # class 0 and class 1
y0 = y[0:100] # class 0 and class 1
nn = NeuralNetMLP(n_output=1,
n_features=X0.shape[1],
n_hidden=10,
l2=0.0,
l1=0.0,
epochs=10,
eta=0.01,
minibatches=10,
shuffle=True,
random_state=1)
nn.fit(X0, y0)
y_pred = nn.predict(X0)
acc = np.sum(y0 == y_pred, axis=0) / float(X0.shape[0])
#print(y_pred)
assert(round(acc, 2) == 1.0)
n_features=X.shape[1],
n_hidden=50,
l2=0.00,
l1=0.0,
epochs=300,
eta=0.01,
alpha=0.0,
decrease_const=0.0,
minibatches=1,
shuffle_init=False,
shuffle_epoch=False,
random_seed=1,
print_progress=3)
nn = nn.fit(X, y)
y_pred = nn.predict(X)
acc = np.sum(y == y_pred, axis=0) / X.shape[0]
print('Accuracy: %.2f%%' % (acc * 100))
plot_decision_regions(X, y, clf=nn, legend=2)
plt.title("Logistic regression - gd")
plt.show()
plt.plot(range(len(nn.cost_)), nn.cost_)
plt.ylim([0, 300])
plt.xlabel('Epochs')
plt.ylabel('Cost')
plt.show()
n_hidden = 50,
l2 = 0.00,
l1 = 0.0,
epochs = 300,
eta = 0.01,
alpha = 0.0,
decrease_const = 0.0,
minibatches = 1,
shuffle_init = False,
shuffle_epoch = False,
random_seed = 1,
print_progress = 3)
nn = nn.fit(X, y)
y_pred = nn.predict(X)
acc = np.sum(y == y_pred, axis=0) / X.shape[0]
print('Accuracy: %.2f%%' % (acc * 100))
plot_decision_regions(X, y, clf=nn, legend=2)
plt.title("Logistic regression - gd")
plt.show()
plt.plot(range(len(nn.cost_)), nn.cost_)
plt.ylim([0, 300])
plt.xlabel('Epochs')
plt.ylabel('Cost')
plt.show()
