simple_classification.py

#!/usr/bin/python
# -*- coding: utf-8 -*-

"""
参考 http://hi-king.hatenablog.com/entry/2015/06/27/194630
クラスタリングの実験
"""

import numpy as np
import chainer as chnr
from chainer import optimizers
import chainer.functions as F
import matplotlib.pyplot as plt

class SmallClassificationModel(chnr.FunctionSet):
    def __init__(self, dsize, lsize):
        # dsize:入力ノード数, lsize:出力ノード数
        super(SmallClassificationModel, self).__init__(
            fc1 = F.Linear(dsize, dsize),
            fc2 = F.Linear(dsize, lsize)
        )
        
    def _forward(self, x):
        #二層以上かつシグモイド関数を使うところが革新的
        h = F.sigmoid(self.fc1(x))
        y = self.fc2(h)
        return y
    
    def train(self, x_data, y_data):
        # データ, ラベルの入力
        x = chnr.Variable(x_data.reshape(1, x_data.shape[0]).astype(np.float32), volatile=False)
        y = chnr.Variable(y_data.astype(np.int32), volatile=False)
        
        # ネットワークに入力した結果
        h = self._forward(x)
        
        # 結果の評価
        optimizer.zero_grads()
        error = F.softmax_cross_entropy(h, y)
        error.backward()
        optimizer.update()

        return h.data.argmax(), error.data

datas_and_labels = [
    [np.array([0,0]),np.array([0])],
    [np.array([0,1]),np.array([1])],
    [np.array([1,0]),np.array([1])],
    [np.array([1,1]),np.array([0])],
]*10000
data_size,label_size = 2, 2  
"""
datas_and_labels = [
    [np.array([0,0,0]),np.array([0])],
    [np.array([1,0,0]),np.array([1])],
    [np.array([0,1,0]),np.array([0])],
    [np.array([0,0,1]),np.array([1])],
    [np.array([1,1,0]),np.array([0])],
    [np.array([1,0,1]),np.array([1])],
    [np.array([0,1,1]),np.array([0])],
    [np.array([1,1,1]),np.array([1])],
]*1000
data_size,label_size = 3, 2  
"""
"""
datas_and_labels = [
    [np.array([0,0,0,0]),np.array([0])],
    [np.array([1,0,0,0]),np.array([1])],
    [np.array([0,1,0,0]),np.array([1])],
    [np.array([0,0,1,0]),np.array([1])],
    [np.array([0,0,0,1]),np.array([1])],
    [np.array([1,1,0,0]),np.array([2])],
    [np.array([1,0,1,0]),np.array([2])],
    [np.array([1,0,0,1]),np.array([2])],
    [np.array([0,1,1,0]),np.array([2])],
    [np.array([0,1,0,1]),np.array([2])],
    [np.array([0,0,1,1]),np.array([2])],
    [np.array([1,1,1,0]),np.array([3])],
    [np.array([1,1,0,1]),np.array([3])],
    [np.array([1,0,1,1]),np.array([3])],
    [np.array([0,1,1,1]),np.array([3])],
    [np.array([1,1,1,1]),np.array([4])],
]*1000     
data_size,label_size = 4, 5  
"""

model = SmallClassificationModel(data_size, label_size)
optimizer = optimizers.MomentumSGD(lr=0.01, momentum=0.9)
optimizer.setup(model.collect_parameters())

errors = []
for epoch,(data, label) in enumerate(datas_and_labels):
    p_class, error = model.train(data, label)
    errors.append(error)
    print("---epoch {}---".format(epoch))
    print("source  label: {l}, data: {d}".format(l=label[0], d=data))    
    print("predict label: {p}, error: {e}".format(p=p_class, e=error))

plt.plot(errors, "b-")
plt.show()