How to create universal type.raw and type_map.raw files?

[IainGross]

I have been pulling data from a variety of runs, which have different atom counts and species, and I want to combine all of my data into a single folder. What I find is that DeepMD does not have any parameter that allows me to shuffle the order in which data is read, so I want to arrange the sets/frames within a single parent folder in a shuffled order. If every .npy binary set must have corresponding type_map.raw and type.raw files that match their atom count and species, how do I organize the data into a single folder? The folder can only have one type.raw and one type_map.raw file, and it seems to me that it is not possible to create "universal" type raw files that account for every .npy set in the dataset when each run has differing atom counts and species.

[wanghan-iapcm]

deepmd-kit randomly samples from systems by the rule explained here, and within each system, data frames are randomly sampled.