data.csv

perceptron_data_view.py

import numpy as np
from sklearn.datasets import make_blobs
from tqdm import tqdm
from matplotlib import colors, pyplot as plt

def sigmoid(z):
    return 1/(1+np.exp(-z))

def perceptron(pesos, data, activ=sigmoid):
    return activ(
            np.sum(pesos*data, axis=1)
            )

def mse(z, target):
    linhas = z.shape[0]
    
    if linhas != target.shape[0]:
        raise "quantidade diferentes de items"

    return 1/linhas * np.sum((z-target)**2)


load_from_csv = True

resolução = 100
bias = -0.5

if load_from_csv:
    data = np.loadtxt("data.csv", delimiter=",")
    target = np.loadtxt("target.csv", delimiter=",")
else:
    data, target = make_blobs(centers=2)
    data = np.c_[data, np.ones(data.shape[0])*-1]

    np.savetxt("data.csv", data, delimiter=",")
    np.savetxt("target.csv", target, delimiter=",")

xmin = data[:, 0].min() -10
xmax = data[:, 0].max() +10
ymin = data[:, 1].min() -10
ymax = data[:, 1].max() +10

x = np.linspace(xmin, xmax, resolução)
y = np.linspace(ymin, ymax, resolução)
#bias = np.arange(-1 , 1, resolução)

#xx, yy, bb = np.meshgrid(x, y, bias)
xx, yy = np.meshgrid(x, y)


#pesosx, pesosy, pesosb = list(map( lambda x: x.ravel(),(xx, yy, bb)))
pesosx, pesosy = list(map( lambda x: x.ravel(),(xx, yy)))

#del x, y, bias

resultados = np.zeros(pesosx.shape[0])

pbar = tqdm(total=pesosx.shape[0])

#for index, (x, y, b) in enumerate(zip(pesosx, pesosy, pesosb)):
for index, (X, Y) in enumerate(zip(pesosx, pesosy)):
    #pesos = np.array((x, y, b))
    pesos = np.array((X, Y, bias))
    z = perceptron(pesos, data)
    resultados[index] = mse(z, target)
    if index % 10 == 0:
        pbar.update(10)
pbar.close()


cols = 6
fig = plt.figure()
gs = fig.add_gridspec(nrows=4, ncols=cols)

ax0 = fig.add_subplot(gs[:2, :3])
ax0.set_aspect("equal")
ax0.scatter(data[:, 0], data[:, 1], c=target)


ax1 = fig.add_subplot(gs[:2, 3:], projection="3d")

Z = resultados.reshape(xx.shape)
ax1.plot_surface(
        xx, yy, Z,
        norm=colors.Normalize(vmin=Z.min(), vmax=Z.max()),
        cmap=plt.cm.viridis_r, alpha=0.75
        )


_Z = Z.ravel()
_xx = xx.ravel()
_yy = yy.ravel()
minimum, maximum = _Z.argmin(), _Z.argmax()
xminimum, xmaximum = _xx[minimum], _xx[maximum]
yminimum, ymaximum = _yy[minimum], _yy[maximum]

cx = np.linspace(xmaximum * np.random.rand(), xminimum, cols)
cy = np.linspace(ymaximum * np.random.rand(), yminimum, cols)
cz = [mse(perceptron(np.array((X, Y, bias)), data), target) for X, Y in zip(cx, cy)]

area = np.array((
    pesosx, pesosy, np.ones(pesosx.shape)*-1
    )).transpose()

ax1.scatter3D(cx, cy, cz, s=50, c="red", data="12345", depthshade=False)
for index, (_x, _y, _z) in enumerate(zip(cx, cy, cz)):
    ax1.text(_x, _y, _z+0.1, f'{index}')

for i in tqdm(range(cols)):
    ax3 = fig.add_subplot(gs[2, i])
    ax3.set_title(f"ponto: {i}\nx = {cx[i]:.2f} | y = {cy[i]:.2f}\nerro={cz[i]:.5f}")
    ax3.scatter(data[:, 0], data[:, 1], c=target)
    

    pesos = np.array((cx[i], cy[i], bias))
    czz =  perceptron(pesos, area)


    ax3.contourf(xx, yy, czz.reshape(xx.shape), alpha=0.25, cmap="viridis")

    ax4 = fig.add_subplot(gs[3, i])
    ax4.scatter(data[:, 0], data[:, 1], c=target, cmap="RdBu", marker="+",s=5)
    X, Y = data[:, 0] * cx[i], data[:, 1] * cy[i]
    ax4.scatter(X, Y, c=target, cmap="RdBu", marker="x", s=15)


fig.tight_layout()
fig.show()

target.csv

1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
1.000000000000000000e+00
0.000000000000000000e+00