Transformation of b and d1

[drcege]

I saw you always transform $b$ and $d_1$ before computing msle:

b = 1.25**b - 1 + 1e-8
d1 = 1.25**d1 - 1 + 1e-8

What does that mean? I did not notice any explanation in the paper.

[ethancaballero]

I do that to make the grid search (scipy.optimize.brute) for b and d1 be spaced logarithmically.

[drcege]

Thanks, that sounds reasonable. And could you please explain a little bit about what 1.25 and 1e-8 are for?

[ethancaballero]

Adding 1e-8 is to prevent b and d1 from being equal to zero. If d1 equals zero, you get a NaN because d1 is in a denominator.

Now that I think about it, adding 1e-8 to b probably isn't necessary anymore; (an older version of BNSL from before version of BNSL in the paper had b in a denominator, and I never got around to removing the addition of 1e-8 to b in the code)

1.25 is an arbitrary base of an exponential function. 1.25 can replaced with any number that is greater than 1.0 (assuming you also readjust the grid search range and resolution), although I would suggest using 1.25 because 1.25 worked fine for me.