교재의 끝판왕 8장 deep_convnet.py 돌린 결과 - LOPES-HUFS/DeepLearningFromForR GitHub Wiki
이 모델은 20에폭을 돌립니다. 돌린 결과는 아래와 같습니다. 생각보다 오래걸림 그러나 보시면 알겠지만 결과는 대박임!
=== epoch:1, train acc:0.099, test acc:0.1 ===
=== epoch:2, train acc:0.976, test acc:0.975 ===
=== epoch:3, train acc:0.977, test acc:0.982 ===
=== epoch:4, train acc:0.99, test acc:0.984 ===
=== epoch:5, train acc:0.992, test acc:0.987 ===
=== epoch:6, train acc:0.991, test acc:0.98 ===
=== epoch:7, train acc:0.991, test acc:0.991 ===
=== epoch:8, train acc:0.994, test acc:0.984 ===
=== epoch:9, train acc:0.99, test acc:0.99 ===
=== epoch:10, train acc:0.994, test acc:0.99 ===
=== epoch:11, train acc:0.995, test acc:0.993 ===
=== epoch:12, train acc:0.994, test acc:0.993 ===
=== epoch:13, train acc:0.998, test acc:0.991 ===
=== epoch:14, train acc:0.996, test acc:0.989 ===
=== epoch:15, train acc:0.995, test acc:0.984 ===
=== epoch:16, train acc:0.997, test acc:0.992 ===
=== epoch:17, train acc:0.998, test acc:0.993 ===
=== epoch:18, train acc:1.0, test acc:0.99 ===
=== epoch:19, train acc:1.0, test acc:0.991 ===
=== epoch:20, train acc:0.999, test acc:0.991 ===
=============== Final Test Accuracy ===============
test acc:0.9943
Saved Network Parameters!그리고 세미나 중 논란이 되었던 loss 값 그래프
자료 보존 차원에서 윗 그래프 그리는데 사용한 R 코드.
> library(tidyverse)
─ Attaching packages ─────────────────────────────────────────────────────────── tidyverse 1.3.0 ─
✔ ggplot2 3.3.0 ✔ purrr 0.3.4
✔ tibble 3.0.1 ✔ dplyr 1.0.0
✔ tidyr 1.1.0 ✔ stringr 1.4.0
✔ readr 1.3.1 ✔ forcats 0.5.0
─ Conflicts ───────────────────────────────────────────────────────────── tidyverse_conflicts() ─
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()
> loss <- read_csv(file="deep_convnet_loss.csv", col_names=FALSE)
Parsed with column specification:
cols(
X1 = col_double()
)
> loss
# A tibble: 12,000 x 1
X1
<dbl>
1 2.29
2 2.29
3 2.32
4 2.28
5 2.27
6 2.28
7 2.30
8 2.27
9 2.26
10 2.27
# … with 11,990 more rows
> mean(loss$X1)
[1] 0.9099704
> median(loss$X1)
[1] 0.8918075
> plot(loss$X1, type="l")