Problem in mlps.run_rmse_test

[YuliangS]

Hi
I am trying to run the example simulation on zundel. 01~04 are finished, but I meet a problem in "05-validation."
"mlps.run_rdf_test" and "mlps.run_vdos_test" work fine, but "mlps.run_rmse_test" does not work.
Here is the traceback:

Could you help me solve this problems?

[beckobert]

run_rdf_test and run_vdos_test uses output from your CP2K MD simulation using the final C-NNP, run_rmse_test uses output from N2P2.
The FileNotFoundError would suggest that an output file from N2P2 is missing. Log and error files in the predict-000 directory should be helpful in figuring out what went wrong.

[beckobert]

Please, see the official documentation of the N2P2 package: https://compphysvienna.github.io/n2p2/topics/units.html

In principle the library and all applications are agnostic to a specific system of physical units. This means that numeric values in input files are processed unaltered. Hence, it is the user’s responsibility to provide data in a consistent way. For instance, the same length units must be used in training configurations (see configuration file format) and in the definition of symmetry functions (mind that some parameters, e.g. , need to be given in length units).

AML works internally in atomic units and all input files to N2P2 are written in atomic units.

[YuliangS]

I got it. Thank you so much.
Sorry I have to bother you with another question.
For the "aml.score" part, could you teach me how I can extract the raw data of RDF? I want to plot the RDF by myself.

[cschran]

The rdfs are all saved as pickle files which you can load separately and plot to your liking.

If you load the pickle file, you get back the dictionary with all rdfs.

[YuliangS]

Thank you for the information!
But why does the RDF value approach 0 instead of approaching 1?

[cschran]

It is a gas phase molecule.