Creates sequence dataset for RNN

sequence_data_gen.py

def dataset_gen():
  sequence_dict = {}
  alphabet_list = [chr(x) for x in range(65,91)]

  for i in range(len(alphabet_list)-5):
    sequence_dict[i] = alphabet_list[i:i+6]
  
  number_trans = 2000
  df_in_list = [(sequence_dict.get(random.randint(1,9))) for x in range(number_trans)]

  df = pd.DataFrame(df_in_list, columns=['week1', 'week2', 'week3', 'week4', 'week5', 'week6'])

  df['seq_string'] = df.apply(lambda x: ' '.join([str(y) for y in x.values.tolist()]), axis=1)

  lines = df['seq_string'].tolist()

  tokenizer = Tokenizer()
  tokenizer.fit_on_texts(lines)
  sequences = tokenizer.texts_to_sequences(lines)

  sequences = np.array(sequences)
  split_ratio = int(0.8*number_trans)
  X, y = sequences[:, :-1], sequences[:,-1]
  X_test, y_test = sequences[split_ratio:, :-1], sequences[split_ratio:,-1] 

  vocab_size = len(tokenizer.word_index) + 1
  y = to_categorical(y, num_classes=vocab_size)
  y_test = to_categorical(y_test, num_classes=vocab_size)

  seq_length = X.shape[1]
  output_dim = seq_length + 1

  return X, y, X_test, y_test, vocab_size, seq_length, output_dim, lines