About the objective function

[Peggyzz]

Hi,there are some issues i dont understand during reading the code

1. objective function
   In optimizer.py, the first term of PPD-SG's objective function is as follows:
   ‘(1-p_hat)tf.reduce_mean(I_postf.reshape((pred_score-a)**2, (-1, 1)) + p_hatI_negtf.reshape((pred_score-b)**2, (-1, 1)))’
   It's hard for me to understand why you compute mean value for both 'pred-a' and 'pred-b' It seems quite different with the objective function in your article
2. the extensions for multi-classes
   How to modify the PPDSG for multi-class? For example, how to compute phat and Phat
   I would be appreciate if you could say more about this extensions