Data-Driven Conditional Robust Optimization

Authors: Abhilash Reddy Chenreddy, Nymisha Bandi, Erick Delage

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we use simulated and real world data to illustrate the implementation of our approach and compare it against two non-contextual robust optimization benchmark approaches to demonstrate the value of exploiting contextual information in robust optimization. In this section, we illustrate the coverage aspect of the IDCC approach using simulated data. We will further demonstrate the advantage of the CRO problem using a standard risk minimizing portfolio optimization problem.
Researcher Affiliation Academia Abhilash Chenreddy GERAD & Dept. of Decision Sciences HEC Montréal Montréal, Quebec, Canada abhilash.chenreddy@hec.ca Nymisha Bandi Mc Gill University Montréal, Quebec, Canada nymisha.bandi@mcgill.ca Erick Delage GERAD & Dept. of Decision Sciences HEC Montréal Montréal, Quebec, Canada erick.delage@hec.ca
Pseudocode Yes Algorithm 1 Integrated deep cluster-classify with deep K-means
Open Source Code Yes The code can be found on github3. Our code uses the Pytorch implementation from [Goerigk and Kurtz, 2020], which is available online4. 3https://anonymous.4open.science/r/Data-Driven-Conditional-Robust-Optimization-E160/ 4https://github.com/goerigk/RO-DNN
Open Datasets Yes Dataset Our experiments make use of historical data from the US stock market. We collect the adjusted daily closing prices for 70 stocks (as used in [Xu and Cohen, 2018]) coming from 8 different sectors from January 1, 2012, to December 31, 2020, using the Y!Finance s API.
Dataset Splits Yes Given the time series nature of the data, at a given instance, we use 3 years of data to train and the following year as validation to pick the hyperparameters of our model such as learning rate, weight decay, and the optimal number of clusters. We then retrain the model using the 4 years of data to build the final model.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running experiments, such as GPU or CPU models.
Software Dependencies Yes MOSEK Ap S. MOSEK Fusion API for C++ 9.3.20, 2022. URL https://docs.mosek.com/latest/cxxfusion/index.html.
Experiment Setup Yes We used learning rate = 0.01, αK = 0.5, αS = 0.5, β = 0.1 for all the experiments.