search

CR-Gjx / LeakGAN

The codes of paper "Long Text Generation via Adversarial Training with Leaked Information" on AAAI 2018. Text generation using GAN and Hierarchical Reinforcement Learning.
https://arxiv.org/abs/1709.08624
576 stars 180 forks source link

An issue about your code #3

Closed Syrup274 closed 6 years ago

Syrup274 commented 6 years ago

I am a little confused by your code synthetic data with sequence length 40. The following code may replace the positive samples(real data) with LSTM generated ones. generate_samples(sess, target_lstm, BATCH_SIZE, generated_num, positive_file, 0) Since the oracle LSTM does not reload parameters for pkl file, how could you treat the random initialized LSTM generated samples as positive samples? Many thanks!

CR-Gjx commented 6 years ago

At the beginning of the Main.py, I import 2 different oracles. You can see that length-20 oracle initializes by reloading pkl file and length-40 oracle will initialize variable from a random number.