Mastering TensorFlow 1.x

Text generation LSTM in Keras

You can follow along with the code for this section in the Jupyter notebook ch-08b_RNN_Text_Keras.

We implement text generation LSTM in Keras with the following steps:

1. First, we convert all of the data into two tensors, tensor x with five columns since we input five words at a time, and tensor y with just one column of output. We convert the y or label tensor into one-hot encoded representation.

Remember that in practice for large datasets you would use word2vec embeddings instead of one-hot representations.

# get the datax_train, y_train = text8.seq_to_xy(seq=text8.part['train'],n_tx=n_x,n_ty=n_y)# reshape input to be [samples, time steps, features]x_train = x_train.reshape(x_train.shape[0], x_train.shape[1],1) ...