Texel tuning: is tanh viable?

[bctboi23]

I was getting back into chess programming and working through a tuning method like the Texel method to try and squeeze out some more elo on the vice clone I have before I go for a fully new engine, and was wondering if it would be a good idea to use the tanh function instead of the sigmoid function, anyone have any experience trying it? My first guess would be that it would perform better than the sigmoid function on tuning, since the gradient near 0 (where most of the tuning gains would likely be had) is larger, but I could be wildy off base.

[hgm]

They are the same, not? Apart from shifting and scaling that makes one range from 0 to 1, instead of -1 to 1.

[bctboi23]

Yeah, essentially they are the same, except for the shifting means it is zero mean. The way texel tuning is described on the chessprogramming wiki has the scaling constant but no shift, and I guess I'm just asking if shifting to zero mean makes sense, the understanding I have being in other learning problems that the zero mean tends to help speed up convergence