ajive_fig2.py 文件源码

def generate_data_ajive_fig2(seed=None):
    """
    Samples the data from AJIVE figure 2. Note here we use rows as observations
    i.e. data matrices are n x d where n = # observations.

    X_obs, X_joint, X_indiv, X_noise, Y_obs, Y_joint, Y_indiv, Y_noise =
    generate_data_ajive_fig2()
    """
    # TODO: return ndarray instead of matrix
    if seed:
        np.random.seed(seed)

    # Sample X data
    X_joint = np.bmat([[np.ones((50, 50))],
                      [-1*np.ones((50, 50))]])
    X_joint = 5000 * np.bmat([X_joint, np.zeros((100, 50))])

    X_indiv = 5000 * np.bmat([[-1 * np.ones((25, 100))],
                              [np.ones((25, 100))],
                              [-1 * np.ones((25, 100))],
                              [np.ones((25, 100))]])

    X_noise = 5000 * np.random.normal(loc=0, scale=1, size=(100, 100))

    X_obs = X_joint + X_indiv + X_noise

    # Sample Y data
    Y_joint = np.bmat([[-1 * np.ones((50, 2000))],
                       [np.ones((50, 2000))]])
    Y_joint = np.bmat([np.zeros((100, 8000)), Y_joint])

    Y_indiv_t = np.bmat([[np.ones((20, 5000))],
                        [-1 * np.ones((20, 5000))],
                        [np.zeros((20, 5000))],
                        [np.ones((20, 5000))],
                        [-1 * np.ones((20, 5000))]])

    Y_indiv_b = np.bmat([[np.ones((25, 5000))],
                        [-1 * np.ones((50, 5000))],
                        [np.ones((25, 5000))]])

    Y_indiv = np.bmat([Y_indiv_t, Y_indiv_b])

    Y_noise = np.random.normal(loc=0, scale=1, size=(100, 10000))

    Y_obs = Y_joint + Y_indiv + Y_noise

    # TODO: make this into a list of dicts i.e. hierarchical
    return X_obs, X_joint, X_indiv, X_noise, Y_obs, Y_joint, Y_indiv, Y_noise