Sample titles from the model

sample_titles.py

# Sample title from the trained model
def sample():   
    num_words = 10
    
    # Initialize input step and hidden state
    input = torch.zeros(1, 1, vocab_size)
    hidden = (torch.zeros(1, 1, n_hidden).to(device), torch.zeros(1, 1, n_hidden).to(device))
    
    i = 0
    output_word = None
    sentence = []
    
    # Sample words from the model
    while output_word != '.' and i < num_words:
          
        input = input.to(device)
        output, next_hidden = rnn(input[0], hidden)
        
        y = output.clone()
        y = y.to(device)
        
        # Use the probabilities from the output to choose the next word
        idx = np.random.choice(range(vocab_size), p = f.softmax(y, dim=1).detach().cpu().numpy().ravel())
        
        output_word = [k for (k, v) in vocab.items() if v == idx][0]
        sentence.append(output_word)
         
        hidden = next_hidden
        input = wordToTensor(output_word)
        i = i+1
        
    return sentence

# Sample 15 titles and print
for i in range(15):
    sampled_title = sample()
    title = ' '.join(sampled_title)
    print(title)
    print("\n")