Obtaining results provided in the original paper

[enekoatxa]

Hi, I have been using the code provided in this repository for some time. The code is brilliant, well structured and easy to understand. However, I am quite far from achieving the results that the original authors of the paper provide.

In the original paper, the results on LaFAN1 dataset are as follows:
L2Q: 0.17, 0.42, 0.69, 0.94 for transition lengths 5, 15, 30 and 45.
L2P: 0.23, 0.65, 0..28, 2.24 for transition lengths 5, 15, 30 and 45.

However, in my attempts, using the hyperparameters provided in the config file (I also tried changing some of them, but i got worse results), I get L2Q values 4.55, 4.61, 4.61 and L2P values 1.79, 2.43, 3.14 for transition lengths 5, 15 and 30.

These values are extremely different from the original values provided. Is my evaluation code wrong? I mean, is it possible to get the values near to the ones in the original article?

Thanks in advance.