CNN 심플넷 구현 - LOPES-HUFS/DeepLearningFromForR GitHub Wiki
구현에 필요한 함수와 라이브러리를 불러와 줍니다. 아래의 소스 코드들을 사용하기 위해 R 환경을 setwd()로 맞추어 줍니다.
library(dslabs)
source("./layers.R")
source("./utils.R")
source("./optimizer.R")
필요한 함수를 불러왔으면 학습에 사용할 데이터를 불러옵니다. 아래 함수를 통해 데이터를 불러옵니다.
init <- function(tensor){
mnist_data <- get_data(tensor)
#손글씨 데이터
x_train_normalize <<- mnist_data$x_train
x_test_normalize <<- mnist_data$x_test
#정답 레이블
t_train_onehotlabel <<- making_one_hot_label(mnist_data$t_train,60000, 10)
t_test_onehotlabel <<- making_one_hot_label(mnist_data$t_test,10000, 10)
}
init(TRUE)
이제 학습할 파라미터를 만들어 줍니다. 파라미터의 개수는 총 6개로 W1, W2, W3, b1, b2, b3 입니다. 아래의 코드를 통해 각 파라미터를 만들 수 있습니다.
simple_net_params <- function(params){
input_size <- params["input_dim"](/LOPES-HUFS/DeepLearningFromForR/wiki/"input_dim")[1]
conv_output_size <- (input_size - params["filter_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_size") + 2*params["pad"](/LOPES-HUFS/DeepLearningFromForR/wiki/"pad")) / params["stride"](/LOPES-HUFS/DeepLearningFromForR/wiki/"stride") + 1
pool_output_size <- as.numeric(params["filter_num"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_num") * (conv_output_size/2)^2)
W1 <- params["weight_init_std"](/LOPES-HUFS/DeepLearningFromForR/wiki/"weight_init_std")*array(rnorm(n = params["input_dim"](/LOPES-HUFS/DeepLearningFromForR/wiki/"input_dim")[3]*params["filter_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_size")^2*params["filter_num"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_num")),dim = c(params["filter_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_size"),params["filter_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_size"),params["input_dim"](/LOPES-HUFS/DeepLearningFromForR/wiki/"input_dim")[3],params["filter_num"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_num")))
b1 <- matrix(rep(0,params["filter_num"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_num")),nrow=1,ncol=params["filter_num"](/LOPES-HUFS/DeepLearningFromForR/wiki/"filter_num"))
W2 <- params["weight_init_std"](/LOPES-HUFS/DeepLearningFromForR/wiki/"weight_init_std")*matrix(rnorm(n = pool_output_size*params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size")),
nrow = pool_output_size, ncol = params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size"))
b2 <- matrix(rep(0,params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size")),nrow=1,ncol=params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size"))
W3 <- params["weight_init_std"](/LOPES-HUFS/DeepLearningFromForR/wiki/"weight_init_std") * matrix(rnorm(params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size")*params["output_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"output_size")),
nrow=params["hidden_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"hidden_size"),ncol=params["output_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"output_size"))
b3 <- matrix(rep(0,params["output_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"output_size")),nrow=1,ncol=params["output_size"](/LOPES-HUFS/DeepLearningFromForR/wiki/"output_size"))
network <- list(W1=W1,b1=b1,W2=W2,b2=b2,W3=W3,b3=b3)
return(network)
}
params <- list(input_dim=c(28,28,1),filter_size=5,filter_num=30,
pad=0,stride=1,hidden_size=100,output_size=10,
weight_init_std=0.01)
network <- simple_net_params(params = params)
모델 학습에 필요한 함수들을 정의해 줍니다.
pool_params <- list(pool_h=2, pool_w=2, stride=2, pad=0)
model.forward <- function(network, x){
conv_params <- list(W = network$W1,b = network$b1, stride = 1, pad = 0)
affine_params_1 <- list(W = network$W2,b = network$b2)
affine_params_2 <- list(W = network$W3,b = network$b3)
conv_temp <- forward("Convolution",x,conv_params)
relu_temp_1 <- forward("ReLU",conv_temp$out)
pool_temp <- forward("Pooling",relu_temp_1$out,pool_params)
flatten_temp <- forward("Flatten",pool_temp$out)
affine_temp_1 <- forward("Affine",flatten_temp$out,affine_params_1)
relu_temp_2 <- forward("ReLU",affine_temp_1$out)
affine_temp_2 <- forward("Affine",relu_temp_2$out,affine_params_2)
return(list(x = affine_temp_2$out, Affine_1.forward = affine_temp_1, Affine_2.forward = affine_temp_2, Relu_1.forward = relu_temp_1,Relu_2.forward = relu_temp_2,conv_temp = conv_temp,pool_temp=pool_temp,flatten = flatten_temp))
}
loss <- function(model.forward, network, x, t){
softmax_params <- list(t = t)
temp <- model.forward(network, x)
y <- temp$x
last_layer.forward <- forward("SoftmaxWithLoss",y,softmax_params) # 이부분 뭔가 이상
return(list(loss = last_layer.forward$loss, softmax = last_layer.forward, predict = temp))
}
model.backward <- function(model.forward, network, x, t) {
# 순전파
loss_temp <- loss(model.forward, network, x, t)
# 역전파
dout <- 1
dout <- backward("SoftmaxWithLoss",loss_temp$softmax,dout)
dout1 <- backward("Affine",loss_temp$predict$Affine_2.forward,dout$dx)
dout2 <- backward("ReLU",loss_temp$predict$Relu_2.forward,dout1$dx)
dout3 <- backward("Affine",loss_temp$predict$Affine_1.forward,dout2$dx)
dout_3 <- backward("Flatten",loss_temp$predict$flatten,dout3$dx)
dx <- backward("Pooling",loss_temp$predict$pool_temp,dout_3,pool_params)
dx1 <- backward("ReLU",loss_temp$predict$Relu_1.forward,dx)
dx2 <- backward("Convolution",loss_temp$predict$conv_temp,dx1$dx)
grads <- list(W1 = dx2$dW, b1 = dx2$db, W2 = dout3$dW, b2 = dout3$db, W3 = dout1$dW, b3 = dout1$db)
return(grads)
}
학습 과정을 저장할 변수를 정의 합니다.
train_loss_list <- data.frame(loss_value = 0)
test_acc <- data.frame(acc = 0)
train_acc <- data.frame(acc = 0)
모델 성능을 평가하는 함수와 데이터를 통한 훈련 과정을 함수로 구현합니다.
model.evaluate <- function(model, network, x, t){
temp <- model(network, x)
y <- max.col(temp$x)
t <- max.col(t)
accuracy <- ifelse(NROW(dim(x))==2,sum(ifelse(y == t,1,0)) / dim(x)[1],
sum(ifelse(y == t,1,0)) / dim(x)[4])
return(accuracy)
}
model.train <- function(train_x,train_t, test_x, test_t, batch_size, epoch, optimizer_name, debug=FALSE){
train_size <- dim(train_x)[4]
iter_per_epoch <- max(train_size / batch_size)
iters_num <- iter_per_epoch*epoch
params <- list(input_dim=c(28,28,1),filter_size=5,filter_num=30,
pad=0,stride=1,hidden_size=100,output_size=10,
weight_init_std=0.01)
network <- simple_net_params(params = params)
for(i in 1:iters_num){
batch_mask <- sample(train_size ,batch_size)
x_batch <- train_x[,,,batch_mask]
x_batch <- array(x_batch,c(28,28,1,100))
t_batch <- train_t[batch_mask,]
gradient <- model.backward(model.forward,network, x_batch, t_batch)
network <- get_optimizer(network, gradient, optimizer_name)
if(debug){
if(i %% iter_per_epoch == 0){
train_acc <- rbind(train_acc ,model.evaluate(model.forward, network, train_x, train_t))
test_acc <- rbind(test_acc, model.evaluate(model.forward, network, test_x, test_t))
train_loss_list <- rbind(train_loss_list,loss(model.forward=model.forward, network = network, x_batch, t_batch)$loss)
}
}
}
return(list(loss=train_loss_list, train_acc=train_acc, test_acc=test_acc))
}
5에폭을 훈련시켜 성능을 평가합니다.
model_5epoch <- model.train(train_x = x_train_normalize,train_t = t_train_onehotlabel,x_test_normalize,t_test_onehotlabel,batch_size = 100, epoch = 5,optimizer_name = "adam",debug = TRUE)