Imperfect information games (IIGs) are a popular subject in the field of artificial intelligence. In this study, we consider them and propose that they can be classified according to the impact and visualizability of the imperfect information. We use Geister, a Board IIG, to create multiple variant games that we use as an abstraction for IIGs. We then train agents to play each variant using deep regret minimization with advantage baselines and model-free learning, a neural-network variation of counterfactual regret minimization. We observe the performance of our agents and use them to qualitatively assess the characteristics of our IIGs with regards to our proposed terminology.