I have a question regarding the data set normalization.
I use the keyword "normalize_data_set" (mode "force").
The energy and force RMSEs obtained are highest than those obtained without normalization.
I executed several training runs with different number of cores, to see how the training results are affected and select the optimal number of cores, and I noticed that the parameters conv_energy and conv_length vary depending on the number of cores.
I don't understand why the number of cores influences the data set normalization... maybe it is related to the fact that normalization is based on predicted forces ?
I would be happy to have more explanation on this subject !
Could you also explain to me what is the role of the keyword "Parallel_Mode" ? Both modes (0 or 1) seem to give similar results.
Hello,
I have a question regarding the data set normalization. I use the keyword "normalize_data_set" (mode "force"). The energy and force RMSEs obtained are highest than those obtained without normalization. I executed several training runs with different number of cores, to see how the training results are affected and select the optimal number of cores, and I noticed that the parameters conv_energy and conv_length vary depending on the number of cores. I don't understand why the number of cores influences the data set normalization... maybe it is related to the fact that normalization is based on predicted forces ? I would be happy to have more explanation on this subject !
Could you also explain to me what is the role of the keyword "Parallel_Mode" ? Both modes (0 or 1) seem to give similar results.
Thank you in advance for your help !