neuron_test.py

"""
Create tests for the neuron

Usage:
python3 neuron_test.py <input_number> <bit_width> <n_tests> <precision> <samples>

Output format:
The file has the following format (example with 4 inputs, i_n n=1,2,3,4):
<i_1> <i_2> <i_3> <i_4> <w_1> <w_2> <w_3> <w_4> <b> <o>
where i stands for input, w for weight, b for bias and o for output
"""

import argparse
import random
from decimal import Context, Decimal, setcontext
from math import ceil, log2

import numpy as np
from test_module import write_dat  # noqa: E402

random.seed(7357)

parser = argparse.ArgumentParser(description="Testing script for neuron")
parser.add_argument("-inp", help="Number of inputs", type=int)
parser.add_argument("-bw", help="Bit width", type=int)
parser.add_argument("-prec", help="Precision", type=int)
parser.add_argument("-tests", help="Number of tests", type=int)
args = parser.parse_args()

RANGE = int(2 ** (args.bw / 2) - 1)
DELAY = 3 + ceil(log2(args.inp))

# Context for Decimal
ctx = Context(prec=args.prec)
setcontext(ctx)

# Test inputs without zero padding
test_inputs = [random.choices(range(-RANGE, RANGE), k=args.inp) for _ in range(args.tests)]

weights = [random.choices(range(-RANGE, RANGE), k=args.inp) for _ in range(args.tests)]

biases = [random.choices(range(-RANGE, RANGE), k=1) for _ in range(args.tests)]

# Create numpy arrays for the calculations
test_inputs = np.array(test_inputs, dtype=Decimal)
weights = np.array(weights, dtype=Decimal)
biases = np.array(biases, dtype=Decimal)

results = np.zeros((args.tests, 1))

# Calculate results
results[:, 0] = np.right_shift(
    np.sum(test_inputs[:, 0 : args.inp] * weights, axis=1) + biases[:, 0], args.prec
)

# Convert arrays back to lists
test_inputs = test_inputs.tolist()
weights = weights.tolist()
biases = biases.tolist()
results = results.tolist()

# Zeros after the computation for the test inputs, weights and biases
for _ in range(DELAY):
    test_inputs.append([0 for _ in range(args.inp)])
    weights.append([0 for _ in range(args.inp)])
    biases.append([0 for _ in range(1)])

# Results with zero padding
results = np.zeros((DELAY, 1)).tolist() + results

write_dat("neuron.dat", args.tests + DELAY, test_inputs, weights, biases, results)