DND 5E CHARACTER BUILDER EQUIPMENT FULL
This imbalance of less than a percent of the full dataset could be problematic later. As it turns out, the Artificer class is a recent addition to the DnD 5e canon, and as such, has not been available long enough for players to choose in significant numbers. The scarcity of Artificer samples in the dataset compared to the other dozen classes is immediately noticeable and a potential cause for concern. A simple question to start: how many of each class are represented in the full dataset? It is important to check if the different categories are well-balanced. Data Explorationįirst, we’ll explore some of the main features of the data and see if there are any surprises that need to be addressed or if there are any discernible patterns in the data to leverage. For now, we’ll ignore these features and equipment too. This could easily lead to overfitting concerns in later stages unless they could be distilled or condensed in a different way. Casting Stat (categorical feature, one of )Īdditional information on the spells and skills available to the characters are provided but will be mostly ignored for a couple reasons: a) many are likely unique to a particular class, and therefore trivialize the classification, b) there are many hundreds of spells and skills in the fully expanded data, which significantly increases the dimensionality for a modestly-sized dataset.Cha(risma), Con(stitution), Dex(terity), Int(elligence), Str(ength), and Wis(dom) Most of this information will not be necessary, as we are primarily interested in the character stats and vitals, namely: The data contains many different features, including user metadata of the submitted entry, as well as the actual character data of interest. The (messy) Jupyter notebook with this analysis can be found in this git repository if you’re interested in following along. I’ll be using python with the pandas and sklearn libraries as the standard toolset.
He lists a few potential caveats with the dataset, none of which seem to be showstoppers for the simple study I am trying to do.
It contains over 7,900 character entries submitted by users via a web application form he created. In order to build a classifier and answer this question, I needed a dataset of DnD characters. In doing this, I began to ponder: how well-balanced is the actual gameplay? Do all optimizations simply lead to the same fundamental character build? Or can I deduce a character’s class based on a limited subset of character information? I wasn’t too keen to read the many sacred tomes that extensively document the game elements to optimize my warlock, but was gently nudged in the right direction by my new DnD compatriots to make the largest stat allocation to Charisma (alas, I have the actual charisma of a sweaty sock to do the role-play justice.) I bungled together a character I thought would be interesting to play - a variant aasimar hexblade warlock. Perhaps it could stave off the monotony that my life had become from thesis-writing in solitary COVID-19 confinement. Not that I was opposed to it, in fact, it sounded fun and I liked that it encouraged collaboration and quick-witted creativity. Despite my respectable nerd cred, I’d never actually played DnD. A couple of months ago, a friend invited me to join him in an online Dungeons and Dragons campaign.