chainerで論理演算を解く

chainer_xor.markdown

概要

chainerで論理演算を学習してとく

env

参照：ほとんど同じ手順、ただしscikit-learnは不要、追加でpipでchainerを入れる https://gist.github.com/masayuki5160/6ab3006bff0e398b1c92ed0f9ded4aee

$ sudo pip install chainer

実行メモ

xor演算はレイヤーが一つのパーセプトロン(xor_fail.py)だとxor演算をうまく解けない。 実行結果はこんな感じ。

('loss:', array(0.2500000596046448, dtype=float32))
('loss:', array(0.2500000596046448, dtype=float32))
[[ 0.4996742 ]
[ 0.49994594]
[ 0.49995193]
[ 0.5002237 ]]

これをレイヤーを増やしてやる(xor_chainer.py)と解けるようになる。

('loss:', array(5.806828085042071e-07, dtype=float32))
[[ 3.73005867e-04]
[ 9.99323368e-01]
[ 9.99324203e-01]
[ 1.12652779e-03]]

or_chainer.py

import numpy as np
import chainer.functions as F
import chainer.links as L
from chainer import Variable,optimizers,Chain

class Model(Chain):
    def __init__(self):
        super(Model, self).__init__(
            l1 = L.Linear(2,1),
            )
    def __call__(self, x):
        # h = self.l1(x)
        # sigmoid function
        h = F.sigmoid(self.l1(x))
        return h

model = Model()
optimizer = optimizers.MomentumSGD(lr=0.01, momentum=0.9)
optimizer.setup(model)

x = Variable(np.array([[0,0],[0,1],[1,0],[1,1]], dtype=np.float32))
t = Variable(np.array([[0],[1],[1],[1]], dtype=np.float32))

for i in range(0,3000):
    optimizer.zero_grads()
    y = model(x)
    loss = F.mean_squared_error(y, t)
    loss.backward()
    optimizer.update()

    print("loss:",loss.data)

print(y.data)

xor_chainer.py

import numpy as np
import chainer.functions as F
import chainer.links as L
from chainer import Variable,optimizers,Chain

class Model(Chain):
    def __init__(self):
        super(Model, self).__init__(
            l1 = L.Linear(2,2),
            l2 = L.Linear(2,1),
            )
    def __call__(self, x):
        # h = self.l1(x)
        # sigmoid function
        h = F.sigmoid(self.l1(x))
        return self.l2(h)

model = Model()
optimizer = optimizers.MomentumSGD(lr=0.01, momentum=0.9)
optimizer.setup(model)

x = Variable(np.array([[0,0],[0,1],[1,0],[1,1]], dtype=np.float32))
t = Variable(np.array([[0],[1],[1],[0]], dtype=np.float32))

for i in range(0,3000):
    optimizer.zero_grads()
    y = model(x)
    loss = F.mean_squared_error(y, t)
    loss.backward()
    optimizer.update()

    print("loss:",loss.data)

print(y.data)

xor_fail.py

import numpy as np
import chainer.functions as F
import chainer.links as L
from chainer import Variable,optimizers,Chain

class Model(Chain):
    def __init__(self):
        super(Model, self).__init__(
            l1 = L.Linear(2,1),
            )
    def __call__(self, x):
        # h = self.l1(x)
        # sigmoid function
        h = F.sigmoid(self.l1(x))
        return h

model = Model()
optimizer = optimizers.MomentumSGD(lr=0.01, momentum=0.9)
optimizer.setup(model)

x = Variable(np.array([[0,0],[0,1],[1,0],[1,1]], dtype=np.float32))
t = Variable(np.array([[0],[1],[1],[0]], dtype=np.float32))

for i in range(0,3000):
    optimizer.zero_grads()
    y = model(x)
    loss = F.mean_squared_error(y, t)
    loss.backward()
    optimizer.update()

    print("loss:",loss.data)

print(y.data)