INNER CODE UNIT · Python

mu1

drawbridge/keras-mmoe · synthetic_demo.py:44

    mu1 = (mu1 - np.mean(mu1)) / (np.std(mu1) * np.sqrt(num_dimension))
    mu2 = np.random.normal(size=num_dimension)
    mu2 -= mu2.dot(mu1) * mu1
    mu2 /= np.linalg.norm(mu2)
    w1 = c * mu1
    w2 = c * (rho * mu1 + np.sqrt(1. - rho ** 2) * mu2)

    # Feature and label generation
    alpha = np.random.normal(size=m)
    beta = np.random.normal(size=m)
    y0 = []
    y1 = []
    X = []

    for i in range(num_row):
        x = np.random.normal(size=num_dimension)
        X.append(x)
        num1 = w1.dot(x)

View source record →