Choice modeling has been a central topic in the study of individual preference or utility across many fields including economics, marketing, operations research, and psychology. While the vast majority of the literature on choice models has been devoted to the analytical properties that lead to managerial and policy-making insights, the existing methods to learn a choice model from empirical data are often either computationally intractable or sample inefficient. In this paper, we develop deep learning-based choice models under two settings of choice modeling: (i) feature-free and (ii) feature-based. Our model captures both the intrinsic utility for each candidate choice and the effect that the assortment has on the choice probability. Synthetic and real data experiments demonstrate the proposed models’ performances in terms of the recovery of the existing choice models, sample complexity, assortment effect, architecture design, and model interpretation. and the maximum depth is tested among 4 , 6 , 8 , 12 , 16 , 18 , 20 . Other parameters are set as default values in the Python package scikit-learn’s function sklearn.ensemble.RandomForestClassifier . We choose the testing structure of our networks following the rule of not getting over complex. We restrict all three parts of our network (product encoder, customer encoder and assort-net) to a depth of no more than 2 . The widths are tuned a little bit loosely (due to their relative lesser effect on the model performance), but no more than 200 . The architecture of TasteNet follows the original paper Han et al. (2020), in which they used 110 hidden units. The architecture of DeepMNL is set to keep some degree of similarity with the architecture of TasteNet.