Open artus-LYTiQ opened 1 month ago
current_entropy = -jnp.sum(current_prob * jnp.log2(jnp.clip(current_prob, 1e-10, 1.0))) is operating on a single probability value. Shouldn't you calculate entropy over a distribution?
where is this code?
current_entropy = -jnp.sum(current_prob * jnp.log2(jnp.clip(current_prob, 1e-10, 1.0))) is operating on a single probability value. Shouldn't you calculate entropy over a distribution?