question about R_hier in Eq(5)

[amiltonwong]

Hi, @diegovalsesia ,

According to the computation on R_hier in hype_triplet_losses, the default option for opposite_hier is False, so this line will be used. From my understanding,

par_norm <--> z_part
p_norm   <--> z_whole

But it is different from the version mentioned in Eq(5) in the original paper. Could you elaborate on it further?

Thanks~

[AntonioMontanaro96]

Hi @amiltonwong ,

sorry for delay, anyway the variables name are a bit misleading:

par_norm <--> parent norm, i.e. z_whole
p_norm <--> positive part norm,i.e. z_part