Dirichlet initialization should take the log of the samples before setting the parameters

HUJI-Deep / simnets-tf

SimNets implementation in TensorFlow

MIT License

7 stars 3 forks source link

Dirichlet initialization should take the log of the samples before setting the parameters #12

Closed orsharir closed 7 years ago

orsharir commented 7 years ago

In your current Dirichlet initialization code, the returned init is simply the samples from the Dirichlet distribution, but it should be the log of the samples instead. See #8 for a fixed initializer code.

elhanan7 commented 7 years ago

Solved in commit 3ab9acb