SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2606.01081 · 2026

Decision-Focused On-Policy Learning for Contextual Linear Optimization with Partial Feedback

Paul Grigas, Wyame Benslimane, Pascal Van Hentenryck, Tinghan Ye

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 8 functions out of this paper's own repositories and ran 6 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
Joeyetinghan/on-policy-bandit-dfl canonical 6 of 8
FunctionStatusWhere it lives
build_lr_schedule Ran Joeyetinghan/on-policy-bandit-dfl/src/algos/lr_schedules.py
code served (permissive licence) · get_code("dc1ae44d02477f4c")
build_nuisance_optimizer Ran Joeyetinghan/on-policy-bandit-dfl/src/common/nuisance_models.py
code served (permissive licence) · get_code("6202d097edebb39a")
compute_log_prob_surrogate Ran Joeyetinghan/on-policy-bandit-dfl/src/common/generative_models.py
code served (permissive licence) · get_code("5aa8934e67b7cbbc")
generative_update_objective Ran Joeyetinghan/on-policy-bandit-dfl/src/common/generative_dfl.py
code served (permissive licence) · get_code("63d84498ef4dc43d")
resolve_torch_device Ran Joeyetinghan/on-policy-bandit-dfl/src/common/generative_models.py
code served (permissive licence) · get_code("f30a6622b4cf953d")
resolve_torch_device Ran Joeyetinghan/on-policy-bandit-dfl/src/common/nuisance_models.py
code served (permissive licence) · get_code("3d143cd1f8994493")
build_nuisance_model Not yet run Joeyetinghan/on-policy-bandit-dfl/src/common/nuisance_models.py
code served (permissive licence) · get_code("eea8f65117d46c59")
create_model Not yet run Joeyetinghan/on-policy-bandit-dfl/src/common/models.py
code served (permissive licence) · get_code("f625dae2acb1036f")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Decision-focused learning (DFL) trains predictive models by optimizing downstream decision quality rather than standalone prediction accuracy. For contextual linear optimization, most existing DFL methods assume offline data and full observations of the objective cost vector. We develop an on-policy learning method for sequential contextual linear optimization under partial feedback, generalizing the standard bandit feedback setting. Our method learns a stochastic predict-thenoptimize policy that samples a cost-vector prediction from a conditional distribution and solves the resulting downstream linear optimization problem. To update this distributional model, we introduce a two-component hybrid gradient estimator. The first component is a score function estimator, which provides an unbiased but potentially high-variance policy gradient estimate. The second is a decision-focused plug-in component that uses an auxiliary nuisance estimate of the latent cost vector to exploit the downstream optimization structure, becoming more informative as the estimate improves. We prove an O(T -1/2 ) bound on the average squared policy-gradient norm, matching the standard non-convex SGD rate. Experiments on top-k selection, shortest path, combinatorial pricing, and a real-data energyscheduling benchmark show that the hybrid gradient approach achieves lower cumulative regret than contextual-bandit-style baselines across all benchmarks, using both Gaussian and richer conditional generative models. Code is available at https://github.com/Joeyetinghan/on-policy-bandit-dfl.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2606.01081")
get_code_for_paper("2606.01081")
have("2606.01081")

Connect an agent — have() is free.