Coordinates density map shows artifact in corners

[Gabriel-p]

The current dataMirror() function does not populate the corners of the frame, resulting in low densities in all four corners. This can be solved with the code shown below.

I changed to perc=.05 because it is a smaller load on memory, with good results. Also, the code below could be made to have better performance by not copying the entire frame six times.

def dataMirror(data, perc=.05):
    """
    Mirror a small percentage of data in all borders to remove the KDE artifact
    that lowers the density close to the edges.

    Source: https://stackoverflow.com/a/33602171/1391441
    """

    # Reshape to: (N_points, 2)
    data = data.T
    # Box
    xmin, ymin = np.min(data, 0)
    xmax, ymax = np.max(data, 0)
    midy = (ymax - ymin) * .5

    points_left_up = np.copy(data)
    points_left_up[:, 0] = xmin - (points_left_up[:, 0] - xmin)
    points_left_up[:, 1] += midy

    points_left_down = np.copy(data)
    points_left_down[:, 0] = xmin - (points_left_down[:, 0] - xmin)
    points_left_down[:, 1] -= midy

    points_right_up = np.copy(data)
    points_right_up[:, 0] = xmax + (xmax - points_right_up[:, 0])
    points_right_up[:, 1] += midy

    points_right_down = np.copy(data)
    points_right_down[:, 0] = xmax + (xmax - points_right_down[:, 0])
    points_right_down[:, 1] -= midy

    points_down = np.copy(data)
    points_down[:, 1] = ymin - (points_down[:, 1] - ymin)
    points_up = np.copy(data)
    points_up[:, 1] = ymax + (ymax - points_up[:, 1])

    # plt.scatter(*data.T, c='g', alpha=.3)
    # plt.scatter(*points_left_up.T, c='b', alpha=.2)
    # plt.scatter(*points_left_down.T, c='r', alpha=.2)
    # plt.scatter(*points_right_up.T, c='b', alpha=.2)
    # plt.scatter(*points_right_down.T, c='r', alpha=.2)
    # plt.scatter(*points_up.T, c='cyan', alpha=.2)
    # plt.scatter(*points_down.T, c='orange', alpha=.2)
    # plt.show()

    # Mirrored data
    mirror_data = np.append(
        np.append(points_left_down, points_left_up, axis=0),
        np.append(points_right_down, points_right_up, axis=0), axis=0)
    mirror_data = np.append(
        mirror_data, np.append(points_down, points_up, axis=0), axis=0)

    # Combine all data
    points = np.append(data, mirror_data, axis=0)

    # Trim mirrored frame to within a 'perc' pad of the full (x, y) range
    xr, yr = np.ptp(data.T[0]) * perc, np.ptp(data.T[1]) * perc
    xmin, xmax = xmin - xr, xmax + xr
    ymin, ymax = ymin - yr, ymax + yr
    msk = (points[:, 0] > xmin) & (points[:, 0] < xmax) &\
        (points[:, 1] > ymin) & (points[:, 1] < ymax)

    return points[msk].T