Concept04_cifar.Rmd

---
title: "Concept04_cifar"
output: github_document
---

학습 정확도 추이 

```{r}
library(ggplot2)
library(data.table)
load("accuracy_list.RData")

accuracy_dt <- rbindlist(accuracy_list)

ggplot(accuracy_dt, aes(epoch, accu)) + geom_point() + 
  geom_line(group=1) + ylab("accuracy") + geom_smooth()

```



학습된 가중치 시각화 

```{r}
source("Concept01_cifar.R")
```


```{r,cache=T}
names_data_labels <- read_data('./cifar-10-batches-py')
```

커널 매트릭스 출력


```{r}
library(tensorflow)
source("Concept03_cnn.R")
source("Concept02_convolution.R")

sess <- tf$InteractiveSession()

saver$restore(sess, "saver/model_1000.chkp-1")

w1_val <- sess$run(W1)

print('weights1:')
show_weights(w1_val)

```



```{r,fig.height=3, fig.width=5}
img_idx <- 222
names_data_labels$names[names_data_labels$labels[img_idx] + 1]
raw_data <- names_data_labels$data[img_idx,]
raw_img <- t(array(raw_data, dim = c(24,24)))

par(mar = rep(1, 4))
image(raw_img, axes = F, col = grey(seq(0, 1, length = 256)))
```

콘볼루션 적용 후 출력 이미지 

```{r}

conv_res <- sess$run(conv_out2, feed_dict=dict(x=matrix(names_data_labels$data[img_idx,], ncol=576)))

print(dim(conv_res))
show_conv_results(conv_res)

sess$close()
```