keras_average.R

library(keras)
df <- read.csv("credit_count.txt")
Y <- matrix(df[df$CARDHLDR == 1, ]$DEFAULT)
X <- scale(df[df$CARDHLDR == 1, ][3:14])

inputs <- layer_input(shape = c(ncol(X)))

mlp <- inputs %>%
  layer_dense(units = 64, activation = 'relu', kernel_initializer = 'he_uniform') %>% 
  layer_dropout(rate = 0.2, seed = 1) %>% 
  layer_dense(units = 64, activation = 'relu', kernel_initializer = 'he_uniform') %>% 
  layer_dropout(rate = 0.2, seed = 1) %>% 
  layer_dense(1, activation = 'sigmoid')

cnv <- inputs %>%
  layer_reshape(c(ncol(X), 1)) %>%
  layer_conv_1d(32, 4, activation = 'relu', padding = "same", kernel_initializer = 'he_uniform') %>%
  layer_max_pooling_1d(2) %>%
  layer_spatial_dropout_1d(0.2) %>%
  layer_flatten() %>%
  layer_dense(1, activation = 'sigmoid')

avg <- layer_average(c(mlp, cnv))
mdl <- keras_model(inputs = inputs, outputs = avg)
mdl %>% compile(optimizer = optimizer_sgd(lr = 0.1, momentum = 0.9), loss = 'binary_crossentropy', metrics = c('binary_accuracy'))
mdl %>% fit(x = X, y = Y, epochs = 50, batch_size = 1000, verbose = 0)
mdl %>% predict(x = X)