altera_devel/scripts/neuron_net_test.py

import numpy as np
from matplotlib import pyplot as plt
from fpu_test_gen import reverse_endian, dtype_size

WEIGTHS = [
    [
        [0.7095146, -0.41895103, -0.08075078, -0.5218736],
        [-0.4351325, 1.0214638, 0.14494987, -0.78134096],
        [0.38553882, -1.0607314, 0.01327306, -0.28972188],
        [0.84955347, 0.32464203, 0.8879888, 0.00756884],
        [-0.8693629, 0.8418823, 0.60206324, -0.78290594],
        [0.1586302, 0.01737848, 0.75329006, -0.57819647],
        [-0.16126093, 0.5317601, 0.34316933, -0.7074082],
        [0.09219088, -0.624525, -0.61903083, -0.87057704]
    ],
    [
        [0.36770403, -0.78046024, 0.3979908, 0.5494289, -0.13859335, 0.40053025, 0.08249452, -0.32528356],
        [-0.17659009, 0.13901198, -0.45248222, -0.7894139, -0.81092286, -0.521815, 0.30632392, -0.3143816],
        [-0.04314173, 0.14361085, 0.6259473, 0.3571782, -0.38011226, 0.01378736, 0.05794358, 0.09667788],
        [-0.46864474, 0.36618456, -0.45595396, -0.39789405, 0.73964316, -0.30294785, 0.2482118, -0.2127953],
        [-0.37941265, 0.45330787, -0.12066315, 0.5636705, 0.68990386, 0.6543718, 0.86367106, -0.5707757],
        [-0.78606385, 0.24032554, -0.4472755, -0.24661142, -0.2698564, -0.8365823, -0.13674814, -0.39799848],
        [0.11138931, 0.48950365, 0.12998834, 0.4947537, 0.516593, 0.82281274, 0.04789656, 0.30206403],
        [0.23097174, 0.30290592, -0.596446, -0.40108407, 0.12246455, -0.47260976, -0.55030185, 0.44481543]
    ],
    [
        [0.5724262, 0.5853241, 0.3748752, -0.892384, -1.0270239, 0.2170913, -0.07271451, 0.14661156],
        [0.30391088, -0.92324615, 0.8088594, -1.0522624, 0.07374455, -0.550893, 0.8194236, -0.62796086]
    ]
]
BIAS = [
    # L1
    [0.01425434, -0.06219335, -0.0201127, -0.04791382, -0.04360008, -0.05311861, -0.01731363, -0.00014839],
    # L2
    [0.03480967, 0.06208326, -0.01576317, -0.00037753, -0.03940378, 0.05157978, -0.02775403, 0.04540931],
    # L3
    [0.03787775, -0.03655371],
]

RESULT = [
    0x00000000, 0x3f800000,
    0x3f800000, 0x00000000,
    0x00000000, 0x00000000,
    0x3f800000, 0x3f800000,
    # 0x3f030126, 0x3f800000,
    # 0x3e652010, 0x00000000,
    # 0x3d2d25da, 0x3efea470,
    # 0x3f1a4267, 0x3f06d0f4,
]


def generate_nn_values(dtype=np.float32):
    dsize = dtype_size(dtype)


def sigm(x):
    ex = 2.7182818 ** x
    return ex / (ex + 1)


def linr(x):
    return x


def process(value):
    if value > 3 or value < 0:
        raise ValueError("Value not in between 0 and 3")
    onehot_mat = [
        [0, 0, 0, 1],
        [0, 0, 1, 0],
        [0, 1, 0, 0],
        [1, 0, 0, 0],
    ]
    L0 = onehot_mat[value]
    L1_mult = [[L0[j] * WEIGTHS[0][i][j] for j in range(len(L0))] for i in range(8)]
    L1 = [linr(sum(L1_mult[i]) + BIAS[0][i]) for i in range(8)]
    L2 = [linr(sum([L1[j] * WEIGTHS[1][i][j] for j in range(len(L1))]) + BIAS[1][i]) for i in range(8)]
    L3 = [sum([L2[j] * WEIGTHS[2][i][j] for j in range(len(L2))]) + BIAS[2][i] for i in range(2)]
    # print(f"L0: {L0} \n\tL1mult: {L1_mult}\n\tL1: {L1} \n\tL2: {L2} \n\tL3: {L3}")
    return [sigm(v) for v in L3]


def show_results():
    sim_data = np.array([process(0), process(1), process(2), process(3)]).T
    hdl_data = np.frombuffer(b''.join([i.to_bytes(4, 'little') for i in RESULT]), dtype=np.float32).reshape(4, 2).T
    print(hdl_data)

    plt.plot(sim_data[0], sim_data[1], 'x', color='b')
    plt.plot(hdl_data[0], hdl_data[1], 'x', color='r')
    plt.xlabel('I')
    plt.ylabel('Q')
    # plt.xlim([0, 1])
    # plt.ylim([0, 1])
    plt.grid()
    plt.show()


if __name__ == '__main__':
    show_results()
    # generate_nn_values()