Hash table second derivitive with respect to the input

[Mehdishishehbor]

Hi,

Thanks for your great research. Does hash table work for second derivative with respect to x (input)? or it only works for weights?

for example, let say we have this sequence of passing info:
x -> hash table -> neural network -> y

Now I want to get the second partial derivative of y with respect to x (d2y/dx2).

Please let me know.

Thanks,
Mehdi

[Tom94]

Hi, yes, this should be supported out of the box.

Importantly, you need to use

"interpolation": "Smoothstep",

when configuring the encoding. When using the default linear interpolation, the second derivative is always zero and hence not very useful.

Please feel free to re-open the issue if you discover a bug in this function.