Express the Calculation of Quadratic Term in Matrix Form

[gtbai]

In line 88-95 of new-FFM.py, when calculating the quadratic term, for now we are using a nested loop, which is super inefficient:

# calculate quadratic term
            self.quad_term = tf.get_variable(name='quad_term', shape=[self.batch_size], dtype=tf.float32)
            for f1 in xrange(0, feature_num - 1):
                for f2 in xrange(f1 + 1, feature_num):
                    W1 = self.quad_weight[f1, self.feature2field[f2]]
                    W2 = self.quad_weight[f2, self.feature2field[f1]]
                    tf.assign_add(self.quad_term, tf.scalar_mul(tf.tensordot(W1, W2, 1), tf.multiply(self.feature_value[:, f1], self.feature_value[:, f2])))

Need to figure out a way to express this in matrix form.