1. Öğren
    2. /
  2. Kurs
    3. /
  3. Recurrent Neural Networks (RNNs) for Language Modeling with Keras
Connected

Egzersiz

Number of parameters comparison

You saw that the one-hot representation is not a good representation of words because it is very sparse. Using the Embedding layer creates a dense representation of the vectors, but also demands a lot of parameters to be learned.

In this exercise you will compare the number of parameters of two models using embeddings and one-hot encoding to see the difference.

The model model_onehot is already loaded in the environment, as well as the Sequential, Dense and GRU from keras. Finally, the parameters vocabulary_size=80000 and sentence_len=200 are also loaded.

Talimatlar

100 XP
  • Import the Embedding layer from keras.layers.
  • On the embedding layer, use vocabulary size plus one as input dimension and sentence size as input length.
  • Compile the model.
  • Print the summary of the model with embedding.