Fashion_MNIST.R

library(tensorflow)
library(keras)
#Sys.setenv(RETICULATE_PYTHON = "./env/Scripts/python.exe")
library(reticulate)


fashion_mnist <- dataset_fashion_mnist()

c(train_images, train_labels) %<-% fashion_mnist$train 
c(test_images, test_labels) %<-% fashion_mnist$test 

dim(train_images)
str(train_images)

fobject <- train_images[5,,]
plot(as.raster(fobject, max = 255))

class_names = c('T-shirt', 'Trousers', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot')

class_names[train_labels[5]+1]

# Normalization

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

#Validation split

val_indices <- 1:5000
val_images <- train_images[val_indices,,]
part_train_images <- train_images[-val_indices,,]
val_labels <- train_labels[val_indices]
part_train_labels <- train_labels[-val_indices]

# Creating Model 

model <- keras_model_sequential()
model %>% 
  layer_flatten(input_shape = c(28, 28)) %>%
  layer_dense(units = 128, activation = 'relu') %>%
  layer_dense(units = 10, activation = 'softmax')

model %>% compile(
  optimizer = 'sgd',
  loss = 'sparse_categorical_crossentropy',
  metrics = c('accuracy')
)
#training

model %>% fit(part_train_images, part_train_labels, epochs = 30,
              batch_size = 100, validation_data = list(val_images, val_labels))

# Test Performance
score <- model %>% evaluate(test_images,test_labels)

cat('Test loss:', score$loss, "\n")
cat('Test accuracy:', score$accuracy, "\n")

# Predicting on Test set

predictions <- model %>% predict(test_images)
predictions[1, ]
which.max(predictions[1, ])
class_names[which.max(predictions[1, ])]
plot(as.raster(test_images[1, , ], max = 1))

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