Calculating Discriminative Score Using TimeGAN

[HsiangHung]

I am using the example code TimeGAN_Synthetic_stock_data.ipynb for timeGAN model.
To train the model and generate time-series data

synth.train(stock_data, train_steps=50000)
synth_data = synth.sample(len(stock_data))

I have two questions:

1. Does synth_data have the same time order as stock_data. For example, are synth_data[123] and synth_data[1011] synthesized from stock_data[123] and stock_data[1011]? I am wondering since in synth.sample(...), ... could be any number. If ... > len(stock_data), how does the index correspondence work?
2. Did anyone calculate the discriminative score using the timeGAN on stock data? I trained a model with 50000 iterations myself, and confirmed all the results shown in the example notebook, except discriminative score. I used GRU or LSTM to build real data/fake data classifier, but always got high accuracy (> 0.9), such that the score > 0.4. According to the paper Time-series Generative Adversarial Networks, discriminative score ~ 0.1