Force loss error

[willphy123]

Hi all,
Does anyone use siesta DFT data to do training? I tested the H2O example. Below is the loss curve diagram. The loss curve is weird. The force error first decreases and then increases to a large value. Any suggestions will be appreciated. Btw, the loss curve is fine when I use VASP DFT data.
Thanks
Will

[njzjz]

In this situation, please read the FAQ and check your data. Since you have two different levels of data, it is much easier to compare them.

[willphy123]

Thanks for your suggestions. Now I solved the problem!