The test results of the pre-trained model are inconsistent with the benchmark about gta HOT 1 CLOSED

Thank you for using our code and providing feedback. It appears that the test data you're using includes CAPE, which is composed of CAPE-NFP and CAPE-FP. In our experiments, we observed that the results on CAPE-FP tend to be better than those on CAPE-NFP. Therefore, your results likely fall between these two. Regarding the differing results in the two papers, we suggest referring to the latest paper. This is due to the fact that in our initial project, we utilized data from ICON, which had some inaccuracies in the CAPE dataset. These inaccuracies were subsequently corrected, leading to potentially improved results with the revised data.

Regarding the normal results, it appears that we inadvertently commented out the following code. This was a testing technique we discovered in ECON's code that can minimize discrepancies. Nevertheless, please note that all models were tested using the same code framework. You can try again, and if you have any problems please contact us!

from gta.