Cleaned up dndbeyond character dataset from kaggle
- Background, feats, inventory items and classes separated into classes themselves
- For classes, a prediction of them being original or customized was made based on the name
- Class has separate flags for the class itself and subclass
- For classes and feats, a "setting" toggle was added, separating "Forgotten Realms" as default and adding
(where possible) a speicifc setting. For unknown "None" is used
- For subclasses found in the wild on reddit or gmbinder, "market" was used as setting
- For inventory items a rough estimation of the "base item" was made for some.
- For inventory items, original name was lost in compression, at some point I'll restore it...
- For inventory items, a gold value listed in the name was transfered into
value_gpfield date_modifiedwas transformed into a datetime object- For classes, an estimation of which subclass belongs to which class was made
- In the dataset itself it's all strings with no relation, so if there was an "Assassin" custom busclass for Fighter class, it would've been missed
- Race was normalized into a class as well, also specifying setting, UA, archived status and customization.
- Where applicable,
__eq__and__hash__were implemented, trying to make it easier to group into sets.- For complex classes a class attribute
SEPARATE_ON_TAGSwere added to switch on and off distinction between same named classes\feats and those customized\UA
- For complex classes a class attribute
For all classes, the same logic persists for those:
setting"Forgotten Realms" for original, "market" for things found on the internet.UAmarked True if specifically mentioned in the name or only exists in UAarchivedmarked True if specifically mentioned in the name or only exists in archived (usually, archived UA)customizedmarked True as an estimation. market things are not marked as such, unless speciifcally mentioned within the name. For those fully custom,setting == None and customized == Trueshould be used