About the REFERENCE dataset

[tszslovewanpu]

Hello and great job!

To reproduce the baselines:
Evaluation:

python scripts/eval.py \
       --ref_path <reference dataset> \
       --gen_path <generated dataset>

What does the reference dataset refer to?

Here's my understanding:

1. Models are trained on the MOSES_trainingset and generate a large number of molecules.
2. The distribution of these generated molecules is compared with the MOSES_testset, which shares the same distribution as the training set.
3. Therefore, the reference dataset serves as a metric to evaluate how well the models have learned the distribution of the MOSES_trainingset.

Is this correct?