fannxy / nfgen Goto Github PK

Hi,

I recently cloned this repo and had a few small issues and questions while trying it out for MP-SPDZ.

1. To get the example.py code working, I had to make a few small changes:
   1. Lines 35 and 36 of the MP-SPDZ fixed-point template at https://github.com/Fannxy/NFGen/blob/main/src/NFGen/CodeTemplet/templet.py use n - I believe they should be m.
   2. I also had to add from Compiler import types to support the @types.vectorize annotation.

With these two changes, MP-SPDZ can compile the generated sigmoid. Is this a correct fix?

2. When I run the code with example.py's default settings, it produces an approximation with a final (k, m) of (6, 8), or 8 pieces of degree-6 polynomials. However, I would like to use it for MP-SPDZ's default fixed-point representation, which is a 31-bit value providing 16 bits of resolution.
   If I adjust the example.py code to use f = 16 and n = 31, leaving other components unchanged, the resulting approximation has many coeffA values that equal 16384.0, but only values on [-16384, 16384) can be converted to fixed-point (not inclusive at the upper bound).
   If I use f = 16, n = 30 instead, the approximation has many -8192.0 and 8192.0 coefficients and uses ~8500 pieces, though it can be compiled under default MP-SPDZ parameters.
   So, I had a few questions:
   1. What would be causing so many maximum coefficients?
   2. Is there a configuration that you would suggest to generate piecewise approximations under the smaller default MP-SPDZ precision?
   3. Or is NFGen designed in such a way that the finer precision is necessary for correct execution and good performance?

Thanks.

Hi, I was wondering if it would be possible to replace the addition, multiplication and comparison protocols designed separately instead of the publicly available library to generate piece-wise polynomial functions for complex functions? If so, how can I do it.

Note on Python 3.8 it works.

Here is the error trace:

ERROR: Could not find a version that satisfies the requirement NFGen (from versions: none)
ERROR: No matching distribution found for NFGen
WARNING: You are using pip version 21.3; however, version 22.3.1 is available.
You should consider upgrading via the '/usr/bin/python3 -m pip install --upgrade pip' command.