November 2019
Intermediate to advanced
346 pages
9h 36m
English
Our initial steps (steps 1, 2, and 4) are operations we have performed during the training phase, which we reproduce here to allow the recipe to be self-contained. In step 3, we read in a seed text to initialize our RNN. The seed text can be any text consisting of the listed characters, as long as it is longer than max_length. Now, we must be able to create interesting text using our pre-trained, pre loaded, and initialized-on-a-seed-text neural network. To this end, we define a convenience function to sample the consequent character that the neural network will generate (step 5). Sampling from the probability vector ensures that the RNN does not simply select the most likely subsequent character, leading to repetitive generated ...