Further changes to preprocess.

      task_ys[task] = np.reshape(np.array(task_ys[task]), (n_samples, 1))

      task_ws[task] = np.reshape(np.array(task_ws[task]), (n_samples, 1))

    return sorted_ids, X, task_ys, task_ws

  elif mode == multitask:

  elif mode == "multitask":

    y = np.zeros((n_samples, n_tasks))

    w = np.ones((n_samples, n_tasks))

    for task in sorted_tasks:

      y[:,task] = np.array(task_ys[task])

      w[:, task] = np.array(task_ws[task])

    for ind, task in enumerate(sorted_tasks):

      y[:,ind] = np.array(task_ys[task])

      w[:,ind] = np.array(task_ws[task])

    return sorted_ids, X, {"all": y}, {"all": w} 

  else:

    raise ValueError("Unsupported mode for process_datasets.")