Rationale of calculating JS

[Puriney]

Hello, I was wondering could you elaborate on the rationale of the way you calculate the JS score.

In the paper, it was splitting the predicted and observed prob into 2 halves and mixed together.

What about

$$JS=KL(\tilde{P_i}|P_i)$$

or

$$JS=KL(P_i|\tilde{P_i})$$

Many thanks.

[Puriney]

KL is asymmetrical (meaning that $$JS=KL(\tilde{P_i}|P_i)$$ and $$JS=KL(P_i|\tilde{P_i})$$ are different JS scores). Using the 'baseline' distribution is the key, therefore, mixing the two in the manuscript seems not appropriate to me. Using the observed spatial expression as the ground truth seems more reasonable to me.