May be wrong

[kidliet]

Thank you for your work！But, I finde some different from the original paper.The "fake_y_next" in your code is generated by "real_x_next" through G，like this：
.In the original code, "fake_y_next" is generated by "fake_y_befor" through the predictor P