gbarp_gen

gbAR(p) process generator in C and Python. Produces correlated binary sequences parameterized by coefficient vector alpha and noise coefficient beta, subject to sum(|alpha|) + beta = 1. Positive coefficients copy the lagged bit; negative coefficients flip it.

Build

./build.sh        # produces build/gbarp_gen and build/libgbarp.so
./build.sh clean

Usage

CLI:

./build/gbarp_gen <output_file> <num_bytes> -a <alpha_vector> [--burn-in <bytes>] [--fast]

-a takes a comma-separated coefficient vector (p inferred from length). -p optionally validates vector length. --burn-in sets thermalization bytes (default 10000). --fast uses xorshift64 instead of /dev/urandom.

./build/gbarp_gen output.bin 1000000 -a 0.3,-0.2,0.1,-0.05
./build/gbarp_gen output.bin 1000000 -a 0.0625,0.0625,0.0625,0.0625 --fast --burn-in 50000

Python:

gbAR(alpha, beta, N_bytes, burn_in_bytes=10000, fast=False) -> bytes

alpha: coefficient vector (list or numpy array). beta: noise coefficient (1 - sum(|alpha|)). N_bytes: output size. burn_in_bytes: thermalization bytes discarded before output. fast: use numpy PRNG instead of os.urandom.

import numpy as np
from gbarp_gen.python import gbAR, constant_alpha

alpha = constant_alpha(p=8, scaling_factor=0.5)
beta = 1 - sum(abs(alpha))
data = gbAR(alpha, beta, N_bytes=100000)

# Negative coefficients flip the copied bit
alpha = np.array([0.2, -0.15, 0.1, -0.05])
beta = 1 - sum(abs(alpha))
data = gbAR(alpha, beta, N_bytes=100000, burn_in_bytes=50000, fast=True)

Alpha family constructors (all return a numpy array):

Function	Parameters	Shape
constant_alpha	p, scaling_factor, signs=None	Equal weights, optional sign pattern
point_to_point_alpha	p, scaling_factor	All weight on lag p
exponentially_decreasing_alpha	p, scaling_factor, decay_rate=1	Exponential decay from lag 1
gaussian_alpha	p, scaling_factor, sigma=None, threshold=0	Bell curve centered at p/2
random_pos_alpha	p, scaling_factor, seed=None	Random positive weights
alternating_sign_alpha	p, scaling_factor	[+, -, +, -, ...] * sf/p

C backend (same signature, optional, faster):

from gbarp_gen.python.c_backend import gbAR_c
data = gbAR_c(alpha, beta, N_bytes=100000)

License

MIT