We consider a preference learning setting where every participant chooses an ordered listofkmost preferred items among adisplayed setofcandidates.

In this work, we examine Asymmetric Shapley Values (ASV), a variant of the popular SHAP additive local explanation method. ASV proposes a way to improve model explanations incorporating known causal relations between variables, and is also considered as a way to test for unfair discrimination in model predictions. Unexplored in previous literature, relaxing symmetry in Shapley values can have counter-intuitive consequences for model explanation. To better understand the method, we first show how local contributions correspond to global contributions of variance reduction. Using variance, we demonstrate multiple cases where ASV yields counter-intuitive attributions, arguably producing incorrect results for root-cause analysis. Second, we identify generalized additive models (GAM) as a restricted class for which ASV exhibits desirable properties. We support our arguments by proving multiple theoretical results about the method. Finally, we demonstrate the use of asymmetric attributions on multiple real-world datasets, comparing the results with and without restricted model families using gradient boosting and deep learning models.

We consider a preference learning setting where every participant chooses an ordered list of $k$ most preferred items among a displayed set of candidates. (The set can be different for every participant.) We identify a distance-based ranking model for the population's preferences and their (ranked) choice behavior. The ranking model resembles the Mallows model but uses a new distance function called Reverse Major Index (RMJ). We find that despite the need to sum over all permutations, the RMJ-based ranking distribution aggregates into (ranked) choice probabilities with simple closed-form expression. We develop effective methods to estimate the model parameters and showcase their generalization power using real data, especially when there is a limited variety of display sets.

Deep reinforcement learning (DRL) has been widely studied in the portfolio management task. However, it is challenging to understand a DRL-based trading strategy because of the black-box nature of deep neural networks. In this paper, we propose an empirical approach to explain the strategies of DRL agents for the portfolio management task. First, we use a linear model in hindsight as the reference model, which finds the best portfolio weights by assuming knowing actual stock returns in foresight. In particular, we use the coefficients of a linear model in hindsight as the reference feature weights. Secondly, for DRL agents, we use integrated gradients to define the feature weights, which are the coefficients between reward and features under a linear regression model. Thirdly, we study the prediction power in two cases, single-step prediction and multi-step prediction. In particular, we quantify the prediction power by calculating the linear correlations between the feature weights of a DRL agent and the reference feature weights, and similarly for machine learning methods. Finally, we evaluate a portfolio management task on Dow Jones 30 constituent stocks during 01/01/2009 to 09/01/2021. Our approach empirically reveals that a DRL agent exhibits a stronger multi-step prediction power than machine learning methods.

In my previous post, we saw that R-squared can lead to a misleading interpretation of the quality of our regression fit, in terms of prediction power. One thing that R-squared offers no protection against is overfitting. On the other hand, cross validation, by allowing us to have cases in our testing set that are different from the cases in our training set, inherently offers protection against overfittting. In this type of validation, one case in our data set is used as the test set, while the remaining cases are used as the training set. We iterate through the data set, until all cases have served as the test set.

In my previous post, we saw that R-squared can lead to a misleading interpretation of the quality of our regression fit, in terms of prediction power. One thing that R-squared offers no protection against is overfitting. On the other hand, cross validation, by allowing us to have cases in our testing set that are different from the cases in our training set, inherently offers protection against overfittting. In this type of validation, one case in our data set is used as the test set, while the remaining cases are used as the training set. We iterate through the data set, until all cases have served as the test set.

In my previous post, we saw that R-squared can lead to a misleading interpretation of the quality of our regression fit, in terms of prediction power. One thing that R-squared offers no protection against is overfitting. On the other hand, cross validation, by allowing us to have cases in our testing set that are different from the cases in our training set, inherently offers protection against overfittting. In this type of validation, one case in our data set is used as the test set, while the remaining cases are used as the training set. We iterate through the data set, until all cases have served as the test set.