Embedding dim in multiclass text classification

Keras
,
1

Hi, I am working on a project where I am trying to predict a pathology from patient anamnesis. The vocab size is 55000, and there are 3 classes.
I use an embedding layer before an LSTM/GRU layer.
What is the best embedding dimension for this case?

2

Hi @Francesca_Pisani .

The documentation of keras.layers.Embedding is here.
In your case, the embedding layer will probably look like this:

tf.keras.layers.Embedding(input_dim=50000, # Number of words in vocabulary
output_dim=EMBEDDING_DIMS, # Dimension of the dense embedding
input_length=MAX_LEN) # Length of the largest input sequences

Note the number of classes (3) is something you’ll take into account in another layer down the road.