def setUp(self):
(self.x_train,
self.y_train), (self.x_test,
self.y_test) = tf.keras.datasets.mnist.load_data()
self.x_train = np.array(self.x_train, dtype=float)
self.x_test = np.array(self.x_test, dtype=float)
self.x_train /= 255
self.x_test /= 255
self.x_train = self.x_train.reshape(784, -1)
self.x_test = self.x_test.reshape(784, -1)
self.y_train = one_hot_encode(self.y_train.reshape(1, -1))
self.y_test = one_hot_encode(self.y_test.reshape(1, -1))
self.x_mini_train = self.x_train[:, :1000].reshape(784, -1)
self.y_mini_train = self.y_train[:, :1000]
self.x_mini_valid = self.x_train[:, 1000:2000].reshape(784, -1)
self.y_mini_valid = self.y_train[:, 1000:2000]
return super().setUp()
def setUp(self):
self.x, self.y = load_iris(return_X_y=True)
self.x = preprocessing.scale(self.x).T
self.y_encoded = one_hot_encode(self.y)
self.softmax_model_no_regularization = SoftmaxRegression(
self.x.shape[0], len(self.y_encoded))
self.softmax_model_l1_regularization = SoftmaxRegression(
self.x.shape[0],
len(self.y_encoded),
regularization="L1",
reg_parameter=0.01)
self.softmax_model_l2_regularization = SoftmaxRegression(
self.x.shape[0],
len(self.y_encoded),
regularization="L2",
reg_parameter=0.01)
self.k_fold_obj = KFoldCrossValidation()
def create_spiral_dataset():
""" Function creates a 2 dimensional spiral dataset, returning
100 feature vectors and encoded and unencoded labels for those
vectors.
"""
num_points = 100 # number of points per class
dimensions = 2 # dimensionality
num_classes = 3 # number of classes
x = np.zeros((num_points * num_classes,
dimensions)) # data matrix (each row = single example)
y = np.zeros(num_points * num_classes, dtype="uint8") # class labels
for j in range(num_classes):
ix = range(num_points * j, num_points * (j + 1))
r = np.linspace(0.0, 1, num_points) # radius
t = np.linspace(
j * 4, (j + 1) * 4,
num_points) + np.random.randn(num_points) * 0.2 # theta
x[ix] = np.c_[r * np.sin(t), r * np.cos(t)]
y[ix] = j
x = x.T
not_encoded_y = y
y = one_hot_encode(y.reshape(1, -1))
return x, y, not_encoded_y