Hands-On Deep Learning with R

Hyperparameter tuning

Let's try changing the embedding_dim hyperparameter to 32 and the batch_size hyperparameter to 50 to see if we get improved results:

# initialize embedding parameterembedding_dim <- 32# train model history <- model %>% fit(  x_train,  y_train,  epochs = 10,  batch_size = 50,  validation_data = list(x_test,y_test),  verbose = 1)# show modelsummary(model)

Here is a printout of the model's architecture:

Now, we will plot the results as follows:

# evaluate resultsplot(history)

This results in the following output:

Unfortunately, ...