Sesion 6

6.1.1 Instalación de los paquetes tensorflow y keras

Primer ejemplo real usando un clasificador de imágenes básicos, usaremos el paquete Tensorflow para predecir y analizar datos de forma más exhaustiva. Usaremos RMarkdown

library(reticulate)

## Warning: package 'reticulate' was built under R version 4.3.3

py_config()

## python:         C:/Users/HP/OneDrive/Documentos/.virtualenvs/r-tensorflow/Scripts/python.exe
## libpython:      C:/Users/HP/AppData/Local/Programs/Python/Python312/python312.dll
## pythonhome:     C:/Users/HP/OneDrive/Documentos/.virtualenvs/r-tensorflow
## version:        3.12.4 (tags/v3.12.4:8e8a4ba, Jun  6 2024, 19:30:16) [MSC v.1940 64 bit (AMD64)]
## Architecture:   64bit
## numpy:          C:/Users/HP/OneDrive/Documentos/.virtualenvs/r-tensorflow/Lib/site-packages/numpy
## numpy_version:  1.26.4
## tensorflow:     C:\Users\HP\OneDrive\DOCUME~1\VIRTUA~1\R-TENS~1\Lib\site-packages\tensorflow\__init__.p
## 
## NOTE: Python version was forced by VIRTUAL_ENV

library(tensorflow)
tf$constant("Hello, TensorFlow!")

## tf.Tensor(b'Hello, TensorFlow!', shape=(), dtype=string)

install.packages("keras")

## Installing package into 'C:/Users/HP/AppData/Local/R/win-library/4.3'
## (as 'lib' is unspecified)

## package 'keras' successfully unpacked and MD5 sums checked
## 
## The downloaded binary packages are in
##  C:\Users\HP\AppData\Local\Temp\Rtmp6pF1xh\downloaded_packages

library(keras) #se pueden hacer estos 2 pasos o ir a la pestaña paquetes, es lo mismo

## Warning: package 'keras' was built under R version 4.3.3

6.1.2 Creando vectores

fashion_mnist <- dataset_fashion_mnist()
#60 mil imágenes de escala grisacea de 28x28px
c(train_images, train_labels) %<-% fashion_mnist$train
#80% de las imágenes
#Arreglo matricial de millones de elementos usados para entrenamiento
#Se usan cuando ya tenemos un resultado y queremos seguir mejorando su visualizando
#o el comportamiento que tienen en el tiempo
c(test_images, test_labels) %<-% fashion_mnist$test
#20% de las imágenes de muestreo
#Con ellas vamos a experimentar el comportamiento previo a un resultado

class_names = c("T-shirt/top", "Trouser", "Pullover", "Dress", "Coat", "Sandal", "Shirt", "Sneaker", "Bag", "Ankle boot")

6.1.3 Explorando datos

dim(train_images) #60000 imágenes calidad de 28x28px 

## [1] 60000    28    28

dim(train_labels) #si hay 60k imágenes hay 60k etiquetas

## [1] 60000

train_labels[1:20]

##  [1] 9 0 0 3 0 2 7 2 5 5 0 9 5 5 7 9 1 0 6 4

#Cada etiqueta es un entero entre 0 a 9, segmentación en cuanto a su arreglo de
#los 20 primeros valores que tiene,
#son 60k imágenes cuyos dígitos ya se encuentran estratificadas y segmentadas de 0 a 9

Hay 10k imágenes en el set de muestra, cada imagen representa una resolución de 28x28px

dim(test_images) #10k imágenes

## [1] 10000    28    28

dim(test_labels) #10k imágenes = 10k etiquetas

## [1] 10000

6.1.4 Construcción del modelo y análisis de datos

library(ggplot2)
library(tidyr)
image_1 <- as.data.frame(train_images[1, , ])
#transformamos un dataframe de la base de datos train_images
colnames(image_1) <- seq_len(ncol(image_1))
image_1$y <- seq_len(nrow(image_1))
#es una variable que se va a ubicar al final de las columnas
image_1 <- gather(image_1, "x", "value", -y)
image_1$x <- as.integer(image_1$x)

ggplot(image_1, aes(x = x, y = y, fill = value)) + 
  geom_tile() +
  scale_fill_gradient(low = "white", high = "black", na.value = NA) +
  scale_y_reverse() +
  theme_minimal() +
  theme(panel.grid = element_blank()) +
  theme(aspect.ratio = 1) +
  xlab(" ") +
  ylab(" ")

train_images <- train_images/255
test_images <- test_images/255

par(mfcol = c(5, 5)) #5 filas, 5 columnas 25 imágenes
par(mar = c(0, 0, 1.5, 0), xaxs = "i", yaxs = "i")
for (i in 1:25){
  img <- train_images[i, , ]
  img <- t(apply(img, 2, rev))
  image(1:28, 1:28, img, col = gray((0:255)/255), xaxt = "n", yaxt = "n",
        main = paste(class_names[train_labels[i] +1]))
}

6.1.5 Elaboración, compilación y entrenamiento del modelo, red neural

Código con versión no compatible actualmente en keras:
library(tidyr)
library(keras)
model <- keras_model_sequential()
model %>%
  layer_flatten(input_shape = c(28, 28)) %>%
  layer_activation('relu') %>%
  layer_activation('softmax')

model %>% compile(
  optimizer = "adam",
  loss = "sparse_categorical_crossentropy",
  metrics = c("accuracy")
)

model %>% fit(train_images, train_labels, epochs = 5, verbose = 5)

score <- model %>% evaluate(test_images, test_labels, verbose = 0)
cat("Test loss:", score[1], "\n")

prediction <- model %>% predict(test_images)
predictions[1, ]

which.max(predictions[1,])

class_pred <- model %>% predict_classes(test_images)
class_pred[1:20]
test_labels[1]

library(keras)
input <- layer_input(shape = c(28, 28))
output <- input %>%
  layer_flatten() %>%
  layer_dense(units = 128, activation = "relu") %>%
  layer_dense(units = 10, activation = "softmax")
model <- keras_model(inputs = input, outputs = output)

model$compile(
  optimizer = 'adam',
  loss = 'sparse_categorical_crossentropy',
  metrics = list('accuracy')
)

epochs <- as.integer(5)
print(class(epochs))

## [1] "integer"

model$fit(train_images, train_labels, epochs = epochs, verbose = 2)

## Epoch 1/5
## 1875/1875 - 6s - 3ms/step - accuracy: 0.8237 - loss: 0.5000
## Epoch 2/5
## 1875/1875 - 4s - 2ms/step - accuracy: 0.8652 - loss: 0.3775
## Epoch 3/5
## 1875/1875 - 4s - 2ms/step - accuracy: 0.8759 - loss: 0.3389
## Epoch 4/5
## 1875/1875 - 5s - 2ms/step - accuracy: 0.8857 - loss: 0.3127
## Epoch 5/5
## 1875/1875 - 5s - 2ms/step - accuracy: 0.8924 - loss: 0.2947

## <keras.src.callbacks.history.History object at 0x000001609E183740>

score <- model$evaluate(test_images, test_labels, verbose = 0)
cat("Test loss:", score[1], "\n")

## Test loss: 0.3552181

predictions <- model$predict(test_images)

## 
##   1/313 ━━━━━━━━━━━━━━━━━━━━ 19s 63ms/step
##  35/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step  
##  67/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
##  98/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 129/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 160/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 193/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 223/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 255/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 288/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step

predictions[1, ]

##  [1] 1.195014e-05 2.915404e-07 9.328386e-07 1.350374e-07 1.844948e-06
##  [6] 3.758254e-03 1.259913e-05 5.831259e-02 1.815648e-05 9.378831e-01

#infitesimales

which.max(predictions[1,]) #la última prenda es la que más se elabora

## [1] 10

predictions <- model$predict(test_images)

## 
##   1/313 ━━━━━━━━━━━━━━━━━━━━ 8s 27ms/step
##  33/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step 
##  68/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 100/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 132/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 165/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 195/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 228/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 264/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 294/313 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step
## 313/313 ━━━━━━━━━━━━━━━━━━━━ 1s 2ms/step

class_pred <- apply(predictions, 1, which.max) - 1
class_pred[1:20]

##  [1] 9 2 1 1 6 1 4 6 5 7 4 5 7 3 4 1 2 2 8 0

test_labels[1]

## [1] 9

6.1.6 Análisis de varias imágenes a la vez

par(mfcol = c(5,5))
par(mar = c(0, 0, 1.5, 0), xaxs = "i", yaxs = "i")
for(i in 1:25){
  img <- test_images[i, , ]
  img <- t(apply(img, 2, rev))
  #subtract 1 as labels go from 0 to 9
  predicted_label <- which.max(predictions[i, ]) -1
  true_label <- test_labels[i]
  if(predicted_label == true_label){
    color <- "#008800"
  } else {
    color <- "#bb0000"
  }
  image(1:28, 1:28, img, col = gray((0:255)/255), xaxt = "n", yaxt = "n",
        main = paste0(class_names[predicted_label + 1], " (",
                      class_names[true_label + 1], ")"),
        col.main = color)
}

img <- test_images[1, , , drop = F]
dim(img)

## [1]  1 28 28

predictions <- model$predict(img)

## 
## 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step
## 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step

predictions

##              [,1]         [,2]         [,3]         [,4]         [,5]
## [1,] 1.195014e-05 2.915399e-07 9.328351e-07 1.350375e-07 1.844945e-06
##             [,6]         [,7]      [,8]        [,9]     [,10]
## [1,] 0.003758245 1.259914e-05 0.0583126 1.81565e-05 0.9378833

prediction <- predictions[1, ] -1
which.max(prediction)

## [1] 10

predictions <- model$predict(img)

## 
## 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 30ms/step
## 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 30ms/step

class_pred <- apply(predictions, 1, which.max) - 1
# Restar 1 para que las clases comiencen desde 0, si es necesario
class_pred

## [1] 9