record one nep.txt every 1e5 generations?

[brucefan1983]

Penghua has suggested to save one nep.txt every 1e5 generations. Then if one trains for 5e5 generations, there will be 5 candidate NEP models that can be used for comparison. These 5 models will be names as e.g., nep100000.txt to nep500000.txt.

Or we can let the user to specify the recording interval?

The main nep.txt will still be updated every 100 generations.

[erhart1]

I think this is a sensible feature. I think the dump frequency should be a parameter that can be specified in the input file. Similarly, it could be useful to add parameters for controlling the frequency for writing RMSEs etc to stdout/loss.out.

[brucefan1983]

This time I only output nep.txt every 100 000 steps (also with date and time info now). I will consider adding more options later.

#316