def linear():
    """
    """
    eigens = np.load('../dataset/v0_eigens.npz')

    train_eigens = eigens['train_eigens']
    issue_eigens = eigens['issue_eigens']

    # 28 boolean a week, labels are the last week
    # train = training, issue = testing, eigens = features
    # it's my convention  :)
    train_labels = train_eigens[:, -28:]
    train_eigens = train_eigens[:, :-28]
    issue_eigens = issue_eigens[:, :-28]

    # flatten the last week (of all users')
    train_labels = np.reshape(train_labels, (np.prod(train_labels.shape), 1))

    train_eigens_temp = []
    issue_eigens_temp = []

    # reshape, there may be some better ways like np.split?
    for i in range(32):
        k = i * 28

        a = train_eigens[:, k:k+28].reshape((train_eigens.shape[0] * 28, 1))
        b = issue_eigens[:, k:k+28].reshape((issue_eigens.shape[0] * 28, 1))

        train_eigens_temp.append(a)
        issue_eigens_temp.append(b)

    train_eigens = np.concatenate(train_eigens_temp, axis=1)
    issue_eigens = np.concatenate(issue_eigens_temp, axis=1)

    # as the picture, assume there are N users in training set:
    # xW = y
    # x: (N * 28) rows * (32 weeks) columns
    # W: (32) rows * (1) column
    # y: (N * 28) rows * (1 week) columns
    model = sklearn.linear_model.LinearRegression()

    model.fit(train_eigens, train_labels)

    guess = model.predict(issue_eigens)

    save_labels(guess)