Data augmentation
Josapha Veloso
09/11/2019
Objetivo
O relatorio é uma descriçao do processo feature engineering com a base de dados feita anteriomente.
O código
suppressMessages(suppressWarnings(library(magick)))
img <- image_read("D:/Users/Usuario/Downloads/RedesNeuraisDigit.jpg")
plot(img)
img <- image_read("D:/Users/Usuario/Downloads/RedesNeuraisDigit.jpg") %>%
image_scale(geometry = 28) %>%
image_data(channels = "gray") %>%
.[1, , ] %>%
as.numeric() %>%
matrix(nrow = 28,ncol = 28,byrow = TRUE) %>%
apply(2,rev) %>% # Pra deixar em pé
t()
img %>%
image() # Serve pra plotar imagens em formato de matriz
Nesse inicio do codigo vamos usar o pacote magik, essencial para fazer ediçoes nas imagens,como assim é feito acima para algumas alteraçoes. A imagem é transformada para a cor cinza e logo após é transformada para um esquema de calor de cores para ser plotada em forma de matriz.
dct28 <- mrbsizeR::dctMatrix(28)
aplica_dct <- function(x){
x <- x/255
q <- x %>%
matrix(28,28,byrow = T) %>%
apply(2,rev) %>%
t()
q <- t(dct28)%*%q%*%dct28
return( as.numeric(q) )
}
img %>%
aplica_dct() %>%
matrix(28,28,byrow = T) %>%
head()## [,1] [,2] [,3] [,4] [,5] [,6]
## [1,] 17.562303 -5.3441382 4.34027006 -2.714131243 1.8284572 -1.12456568
## [2,] -5.076840 1.3756457 -1.40944686 1.057768241 -0.3662663 0.06546695
## [3,] 4.261970 -1.4304730 0.38616292 -0.214611420 0.6644516 -0.42282668
## [4,] -2.256486 0.7778634 -0.53449904 0.007334087 -0.2464381 0.44883571
## [5,] 2.081315 -0.5538839 0.45181465 -0.362750459 0.2623780 -0.06869007
## [6,] -1.536523 0.4252067 0.01021965 0.319901084 -0.3086368 -0.06672195
## [,7] [,8] [,9] [,10] [,11]
## [1,] 1.70012473 -1.178604956 1.18540781 -0.40851739 0.98276068
## [2,] -0.67689381 0.495486495 -0.18232281 0.02605626 -0.44785992
## [3,] 0.34908440 0.002754063 0.08491686 -0.35737702 0.36583473
## [4,] -0.08392305 -0.045254498 -0.28917203 0.11865271 0.02430075
## [5,] 0.07853420 -0.218146598 0.27929773 0.02138392 -0.06201612
## [6,] -0.06589996 0.144141657 -0.03333580 0.08560158 -0.10105244
## [,12] [,13] [,14] [,15] [,16]
## [1,] -0.66169018 0.78866885 -0.24057918 0.71462773 -0.234603970
## [2,] 0.21749439 -0.09130330 0.05055694 -0.33806578 0.085884972
## [3,] 0.10407613 0.02409175 -0.18197662 0.28480082 0.001479248
## [4,] 0.07494815 -0.19737822 0.07251261 0.02360130 -0.052226330
## [5,] -0.15230789 0.27156301 0.02888095 -0.10433168 -0.071594822
## [6,] -0.02273319 -0.03610431 0.01683664 -0.08416796 0.101010088
## [,17] [,18] [,19] [,20] [,21]
## [1,] 0.61246291 -0.15159252 0.45379447 -0.05194074 0.48203513
## [2,] -0.06963867 0.03489345 -0.21862496 0.00724903 -0.08864592
## [3,] -0.04131617 -0.04567576 0.26977685 0.02935309 -0.03508404
## [4,] -0.18781814 0.10424978 0.04535343 -0.06316358 -0.12243171
## [5,] 0.24555265 0.01508982 -0.10144713 -0.02413937 0.19083375
## [6,] 0.02761969 -0.10169223 -0.10767724 0.10472828 0.01427389
## [,22] [,23] [,24] [,25] [,26]
## [1,] -0.0006051326 0.35351339 0.10534565 0.31284031 0.08845171
## [2,] 0.0140463932 -0.12750268 -0.05075131 -0.09126270 -0.01584536
## [3,] -0.0229521257 0.15208364 0.02931851 0.03518324 0.05534414
## [4,] 0.0424074316 -0.02261758 -0.06053511 -0.04370488 0.03602405
## [5,] 0.0087010440 -0.07382461 0.01409086 0.12280976 -0.01413816
## [6,] -0.0872384161 -0.03777471 0.08655306 -0.01040531 -0.10138114
## [,27] [,28]
## [1,] 0.22820877 0.23528821
## [2,] -0.06221214 -0.07847949
## [3,] 0.08591323 0.01373002
## [4,] -0.01094851 -0.07118311
## [5,] -0.03827272 0.06662419
## [6,] -0.05777931 0.06437127dct28 <- mrbsizeR::dctMatrix(28)
aplica_dct <- function(x){
x <- x/255
q <- x %>%
matrix(28,28,byrow = T) %>%
apply(2,rev) %>%
t()
q <- t(dct28)%*%q%*%dct28
return( as.numeric(q) )
}
img %>%
aplica_dct() %>%
matrix(28,28,byrow = T) %>%
head()## [,1] [,2] [,3] [,4] [,5] [,6]
## [1,] 17.562303 -5.3441382 4.34027006 -2.714131243 1.8284572 -1.12456568
## [2,] -5.076840 1.3756457 -1.40944686 1.057768241 -0.3662663 0.06546695
## [3,] 4.261970 -1.4304730 0.38616292 -0.214611420 0.6644516 -0.42282668
## [4,] -2.256486 0.7778634 -0.53449904 0.007334087 -0.2464381 0.44883571
## [5,] 2.081315 -0.5538839 0.45181465 -0.362750459 0.2623780 -0.06869007
## [6,] -1.536523 0.4252067 0.01021965 0.319901084 -0.3086368 -0.06672195
## [,7] [,8] [,9] [,10] [,11]
## [1,] 1.70012473 -1.178604956 1.18540781 -0.40851739 0.98276068
## [2,] -0.67689381 0.495486495 -0.18232281 0.02605626 -0.44785992
## [3,] 0.34908440 0.002754063 0.08491686 -0.35737702 0.36583473
## [4,] -0.08392305 -0.045254498 -0.28917203 0.11865271 0.02430075
## [5,] 0.07853420 -0.218146598 0.27929773 0.02138392 -0.06201612
## [6,] -0.06589996 0.144141657 -0.03333580 0.08560158 -0.10105244
## [,12] [,13] [,14] [,15] [,16]
## [1,] -0.66169018 0.78866885 -0.24057918 0.71462773 -0.234603970
## [2,] 0.21749439 -0.09130330 0.05055694 -0.33806578 0.085884972
## [3,] 0.10407613 0.02409175 -0.18197662 0.28480082 0.001479248
## [4,] 0.07494815 -0.19737822 0.07251261 0.02360130 -0.052226330
## [5,] -0.15230789 0.27156301 0.02888095 -0.10433168 -0.071594822
## [6,] -0.02273319 -0.03610431 0.01683664 -0.08416796 0.101010088
## [,17] [,18] [,19] [,20] [,21]
## [1,] 0.61246291 -0.15159252 0.45379447 -0.05194074 0.48203513
## [2,] -0.06963867 0.03489345 -0.21862496 0.00724903 -0.08864592
## [3,] -0.04131617 -0.04567576 0.26977685 0.02935309 -0.03508404
## [4,] -0.18781814 0.10424978 0.04535343 -0.06316358 -0.12243171
## [5,] 0.24555265 0.01508982 -0.10144713 -0.02413937 0.19083375
## [6,] 0.02761969 -0.10169223 -0.10767724 0.10472828 0.01427389
## [,22] [,23] [,24] [,25] [,26]
## [1,] -0.0006051326 0.35351339 0.10534565 0.31284031 0.08845171
## [2,] 0.0140463932 -0.12750268 -0.05075131 -0.09126270 -0.01584536
## [3,] -0.0229521257 0.15208364 0.02931851 0.03518324 0.05534414
## [4,] 0.0424074316 -0.02261758 -0.06053511 -0.04370488 0.03602405
## [5,] 0.0087010440 -0.07382461 0.01409086 0.12280976 -0.01413816
## [6,] -0.0872384161 -0.03777471 0.08655306 -0.01040531 -0.10138114
## [,27] [,28]
## [1,] 0.22820877 0.23528821
## [2,] -0.06221214 -0.07847949
## [3,] 0.08591323 0.01373002
## [4,] -0.01094851 -0.07118311
## [5,] -0.03827272 0.06662419
## [6,] -0.05777931 0.06437127# Imagem
img %>%
aplica_dct() %>%
matrix(28,28,byrow = T) %>%
apply(2,rev) %>%
t() %>%
image()
Logo após ela é editada para ficar no formato 28x28.
train <- read_rds("~/redesneurais/digit-recognizer/train.rds")
label <- read_rds("~/redesneurais/digit-recognizer/label_test.rds")
set.seed(13092019)
x_train <- train %>%
filter(Flag == 0) %>%
select(-label, -Flag) %>%
mutate_all(function(x) x/255) %>%
as.matrix()
y_train <- train %>%
filter(Flag == 0) %>%
select(label) %>% as.matrix() %>%
keras::to_categorical(num_classes = 10)
# base de treino
x_train_transf <- x_train %>%
apply(1,aplica_dct) %>%
t()Rotacionando a imagem, temos:
aplica_inversao <- function(x,theta){
x %>%
matrix(28,28,byrow = T) %>%
imager::as.cimg() %>%
imager::rotate_xy(theta,15,15) %>%
as.vector() %>%
return()
}
# Rotacao 45º
as.numeric(img) %>% aplica_inversao(45) %>% matrix(28,28,byrow = T) %>% image()
e aplicando a base de treinamento:
# Aplicando à base de dados
x_train_inversao_45 <- x_train %>%
apply(1,aplica_inversao,45) %>%
t()
x_train_inversao_90 <- x_train %>%
apply(1,aplica_inversao,90) %>%
t()
x_train_inversao_45_dct <- x_train_transf %>%
apply(1,aplica_inversao,45) %>%
t()
x_train_inversao_90_dct <- x_train_transf %>%
apply(1,aplica_inversao,90) %>%
t()
# Juntando as matrizes
x_train_max <- cbind(rbind(x_train,x_train_inversao_45,x_train_inversao_90),
rbind(x_train_transf,x_train_inversao_45_dct,
x_train_inversao_90_dct))