Does the order of elements in type_map in the JSON file have to be consistent with the order of type.raw in the initial training dataset?

[wangjiang-j]

Does the order of elements in the type_map in the JSON file have to match the order in type.raw from the initial training dataset? I have high-entropy alloy datasets with different element compositions and orders. Do I need to modify the type.raw of these datasets to be the same?

[iProzd]

Hi @wangjiang-j ,

no, the order of elements in the type_map of the JSON file does not need to match the order in type.raw from the initial training dataset. It will be automatically mapped based on their names, as long as:

1. The type_map of the JSON file includes all elements required for your datasets.
2. Each dataset’s type.raw and type_map.raw maintain internal consistency (i.e., element are correctly defined within files). You can use dpdata to automatically convert from DFT data from other softwares.

So you don't have to modify the mapping of your datasets. Hope this helps :)