5
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 5 years have passed since last update.

SENSYAdvent Calendar 2017

Day 14

Theanoを使って、CNNを実装

Last updated at Posted at 2017-12-14

SENSY Advent Calendar14日目の記事です。

12月からSENSY株式会社のAIチームで働いています。今回はDeep Learningへの応用ができるTheanoという数値計算ライブラリを用いて、チュートリアルを参考に畳み込みニューラルネットワーク(CNN)の実装をしたので、実装内容と結果をまとめました。細かい説明は省いています。

データはMNISTを用いました。

インストール

$ pip install Theano 

エラーなくversion'1.0.1'がインストールができました。
動作確認として、以下のコードを実行します。

from theano import function, config, shared, tensor
import numpy
import time

vlen = 10 * 30 * 768  
iters = 1000

rng = numpy.random.RandomState(22)
x = shared(numpy.asarray(rng.rand(vlen), config.floatX))
f = function([], tensor.exp(x))
print(f.maker.fgraph.toposort())
t0 = time.time()
for i in range(iters):
    r = f()
t1 = time.time()
print("Looping %d times took %f seconds" % (iters, t1 - t0))
print("Result is %s" % (r,))
if numpy.any([isinstance(x.op, tensor.Elemwise) and
              ('Gpu' not in type(x.op).__name__)
              for x in f.maker.fgraph.toposort()]):
    print('Used the cpu')
else:
    print('Used the gpu')

実行結果

[Elemwise{exp,no_inplace}(<TensorType(float64, vector)>)]
Looping 1000 times took 2.734218 seconds
Result is [ 1.23178032  1.61879341  1.52278065 ...,  2.20771815  2.29967753 1.62323285]
Used the cpu

CPUでの動作確認ができました。

MNISTのデータダウンロード

今回はscikit-learnのmldataを用いてダウンロードしました。

from sklearn.datasets import fetch_mldata
mnist = fetch_mldata('MNIST original', data_home='.')
>>> mnint
{'COL_NAMES': ['label', 'data'],
 'DESCR': 'mldata.org dataset: mnist-original',
 'data': array([[0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        ..., 
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0]], dtype=uint8),
 'target': array([ 0.,  0.,  0., ...,  9.,  9.,  9.])}

28×28ピクセルの0~9の手書き数字データがダウンロードできました。

実装

インポート

numpy,theanoなどのライブラリをインストールします。

import numpy as np
import os
import time
import theano
import theano.tensor as T
from theano.tensor.nnet import conv2d
from theano.tensor.signal.pool import pool_2d
from sklearn.model_selection import train_test_split
from sklearn.datasets import fetch_mldata

データロード

データを読み込み、、80%を学習用、20%を評価用と分割して、複数の関数から参照できる共有変数とします。

def data_load(dataset):
    
    mnist = fetch_mldata(dataset, data_home='.')
    train_set_x ,test_set_x, train_set_y,test_set_y = train_test_split(mnist.data,mnist.target, train_size=0.8)
    train_set = [train_set_x,train_set_y]
    test_set = [test_set_x, test_set_y]
    
    def shared_dataset(data, borrow=True):
        data_x, data_y = data
        
        shared_x = theano.shared(
            np.asarray(data_x, dtype=theano.config.floatX), borrow=borrow)
        shared_y = theano.shared(
            np.asarray(data_y, dtype=theano.config.floatX), borrow=borrow)
        
        return shared_x, T.cast(shared_y, 'int32')
    
    test_set_x, test_set_y = shared_dataset(test_set)
    train_set_x, train_set_y = shared_dataset(train_set)
    
    rval = [(train_set_x, train_set_y),
            (test_set_x, test_set_y)]
    
    return rval

畳み込み層

CNNの畳み込み層を実装します。

class ConvLayer(object):
    def __init__(self, rng, input, filter_shape, image_shape):
        assert image_shape[1] == filter_shape[1]
        self.input = input
        fan_in = np.prod(filter_shape[1:])
        fan_out = filter_shape[0] * np.prod(filter_shape[2:])
        
        W_bound = np.sqrt(6.0 / (fan_in + fan_out))
        self.W = theano.shared(
            np.asarray(
                rng.uniform(low=-W_bound, high=W_bound, size = filter_shape),
                dtype=theano.config.floatX
            ),
            borrow=True
        )
        
        b_values = np.zeros((filter_shape[0],),dtype=theano.config.floatX)
        self.b = theano.shared(value=b_values, borrow=True)
        
        conv_out = conv2d(
            input=input,
            filters=self.W,
            filter_shape=filter_shape,
            input_shape=image_shape
        )
        
        self.output = T.tanh(conv_out + self.b.dimshuffle('x', 0, 'x', 'x'))
        
        self.params = [self.W, self.b]

ブーリング層

CNNのプーリング層を実装します。

class PoolLayer(object):
    def __init__(self, rng, input, poolsize=(2, 2)):
        
        pooled_out = pool_2d(
            input = input ,
            ws=poolsize,
            ignore_border=True)
        
        self.output = pooled_out

隠れ層

畳み込み、プーリング後のニューラルネットワーク部分の隠れ層を実装します。

class HiddenLayer(object):
    def __init__(self, rng, input, n_in, n_out, W=None, b=None, activation=T.tanh):
        self.input = input
        
        if W is None:
            W_bound = np.sqrt(6.0 / (n_in + n_out))
            W_values = np.asarray(
                rng.uniform(
                    low=-W_bound, high=W_bound, size = (n_in, n_out)),
                dtype=theano.config.floatX
            )
            if activation == T.nnet.sigmoid:
                W_values += 4
            
            W = theano.shared(value=W_values, name='W', borrow=True)
            
        if b is None:
            b_values = np.zeros((n_out,),dtype=theano.config.floatX)
            b = theano.shared(value=b_values, name='b',borrow=True)
            
        self.W = W
        self.b = b
        
        lin_output = T.dot(input, self.W) + self.b
        self.output = (
            lin_output if activation is None
            else activation(lin_output)
        )
        
        self.params = [self.W, self.b]

ロジスティック回帰

class LogisticRegression(object):

    def __init__(self, input, n_in, n_out):
        self.W = theano.shared(
            value=np.zeros(
                (n_in, n_out),
                dtype=theano.config.floatX),
            name='W',
            borrow=True
        )
        
        self.b = theano.shared(
            value=np.zeros(
                (n_out,),
                dtype=theano.config.floatX),
            name='b',
            borrow=True
        )
        
        self.p_y_given_x = T.nnet.softmax(T.dot(input, self.W) + self.b)
        
        self.y_pred = T.argmax(self.p_y_given_x, axis=1)
        
        self.params = [self.W, self.b]
        
    def negative_log_likelihood(self, y):
        return -T.mean(T.log(self.p_y_given_x)[T.arange(y.shape[0]), y])
    
    def errors(self, y):
        if y.ndim != self.y_pred.ndim:
            raise TypeError('y should have the same shape as self.y_pred',
                            ('y', y.type, 'y_pred', self.y_pred.type))

        if y.dtype.startswith('int'):
            return T.mean(T.neq(self.y_pred, y))
        else:
            raise NotImplementedError()

学習

def lenet(lr=0.1, epoch=100, dataset='MNIST original', batch_size=500):
    rng = np.random.RandomState(23455)
    
    datasets = data_load(dataset)
    train_x, train_y = datasets[0]
    test_x, test_y = datasets[1]
    n_train_batches = int(train_x.get_value(borrow=True).shape[0] / batch_size)
    n_test_batches = int(test_x.get_value(borrow=True).shape[0] / batch_size)
    
    index = T.lscalar()
    x = T.matrix('x')
    y = T.ivector('y')
    
    layer0_input = x.reshape((batch_size, 1, 28, 28))
    
    layer0 = ConvLayer(rng,
                       input = layer0_input,
                       filter_shape = (10, 1, 5, 5),
                       image_shape = (batch_size, 1, 28, 28))
    
    layer1 = ConvLayer(rng,
                       input = layer0.output,
                       filter_shape = (20, 10, 5, 5),
                       image_shape = (batch_size, 10, 24, 24))
    
    layer2 = PoolLayer(rng,
                       input = layer1.output,
                      poolsize=(2, 2))
                       
    
    layer3 = ConvLayer(rng,
                       input = layer2.output,
                       filter_shape = (50, 20, 5, 5),
                       image_shape = (batch_size, 20, 10, 10))
    
    layer4 = PoolLayer(rng,
                       input = layer3.output,
                      poolsize=(2,2))
    
    layer5_input = layer4.output.flatten(2)
    
    layer5 = HiddenLayer(rng,
                         input = layer5_input,
                         n_in = 50 * 3 * 3,
                         n_out = 100,
                         activation = T.tanh)
    
    layer6 = LogisticRegression(input = layer5.output,
                                n_in = 100,
                                n_out = 10)
    
    cost = layer6.negative_log_likelihood(y)
    
    test_model = theano.function(
        [index],
        layer6.errors(y),
        givens={
            x:test_x[index * batch_size: (index + 1) * batch_size],
            y:test_y[index * batch_size: (index + 1) * batch_size]
        })
    
    params = layer6.params + layer5.params + layer3.params + layer1.params + layer0.params    
    grads = T.grad(cost, params)
    updates = [(param_i, param_i - lr * grad_i) for param_i,grad_i in zip(params, grads)]
    
    train_model = theano.function(
        [index],
        cost,
        updates=updates,
        givens={
            x:train_x[index * batch_size: (index + 1) * batch_size],
            y:train_y[index * batch_size: (index + 1) * batch_size]
        })
    
    start_time = time.clock()
    test_score = 0
    
    for epo in range(epoch):
        for minibatch_index in range(n_train_batches):
            cost_i = train_model(minibatch_index)
            
        test_loss = [test_model(i) for i in range(n_test_batches)]
        test_score = np.mean(test_loss)
        print ('epo : %d , test accuracy : %f' % (epo, 100.0 - test_score * 100))
            
    end_time = time.clock() 

今回はConv→Conv→Pool→Conv→Pool→Hidden→Outputという形で実装しました。
batch_sizeを500で100epoch学習しました。
実行結果

test accuracy : 99.22%

output.png

学習にかかった時間 : 310.67m

まとめ

今回はTheanoを用いてCNNを実装し、MNISTのデータを学習しました。
機会があれば、GPUでも動かしてみようと思います。

参考文献

Deep Learning tutorial

5
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
5
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?