threshold-8to3encoder

8-to-3 priority encoder. Outputs binary index of highest-priority set input.

Function

encode(I7..I0) -> (Y2, Y1, Y0)

Priority: I7 > I6 > I5 > I4 > I3 > I2 > I1 > I0

Example Encodings

Input	Highest	Output
10000000	I7	111 (7)
01000000	I6	110 (6)
00100000	I5	101 (5)
00010000	I4	100 (4)
00001000	I3	011 (3)
00000100	I2	010 (2)
00000010	I1	001 (1)
00000001	I0	000 (0)
11111111	I7	111 (7)

Architecture

Single layer with 3 neurons using weighted priority:

Output	Function	Weights [I7..I0]	Bias
Y2	I7∨I6∨I5∨I4	[1,1,1,1,0,0,0,0]	-1
Y1	Priority bit 1	[16,16,-4,-4,1,1,0,0]	-1
Y0	Priority bit 0	[128,-64,32,-16,8,-4,2,0]	-1

Y1 and Y0 use weighted dominance: higher-priority inputs have larger weights that override lower-priority inputs through the threshold mechanism.

Parameters

Inputs	8
Outputs	3
Neurons	3
Layers	1
Parameters	27
Magnitude	303

Usage

from safetensors.torch import load_file
import torch

w = load_file('model.safetensors')

def encode8to3(i7, i6, i5, i4, i3, i2, i1, i0):
    inp = torch.tensor([float(i7), float(i6), float(i5), float(i4),
                        float(i3), float(i2), float(i1), float(i0)])
    y2 = int((inp @ w['y2.weight'].T + w['y2.bias'] >= 0).item())
    y1 = int((inp @ w['y1.weight'].T + w['y1.bias'] >= 0).item())
    y0 = int((inp @ w['y0.weight'].T + w['y0.bias'] >= 0).item())
    return y2, y1, y0

# I5 is highest set bit
print(encode8to3(0, 0, 1, 0, 1, 0, 0, 0))  # (1, 0, 1) = 5

License

MIT