Reinforcement Clustering on Categorical Data - 11/02/2026 18:42 EST
Budget: $50 – $200 USD
I have a fully categorical data set and I want to explore a Reinforcement learning approach to achieve Clustering—steering clear of the usual supervised pipelines. The core goal is to let an agent interact with the data space, discover coherent groups, and maximise a reward function that reflects intra-cluster similarity and inter-cluster separation.
Here is what I need from you:
• Clean, well-commented code (Python preferred) that builds the reinforcement environment around the categorical features, implements the agent, and trains it until a stable set of clusters emerges.
• A brief report or notebook showing how the reward signal is defined, how you tuned hyper-parameters, and how the resulting clusters can be interpreted or visualised.
• Clear instructions so I can reproduce the run on my machine—package versions, command-line calls, and any preprocessing steps.
If you have previous experience combining Reinforcement learning with clustering or handling purely categorical data through embedding tricks, that will be a plus. Source files and reproducibility are the acceptance criteria; once I can retrain and obtain comparable clustering scores, the job is complete.
Here is what I need from you:
• Clean, well-commented code (Python preferred) that builds the reinforcement environment around the categorical features, implements the agent, and trains it until a stable set of clusters emerges.
• A brief report or notebook showing how the reward signal is defined, how you tuned hyper-parameters, and how the resulting clusters can be interpreted or visualised.
• Clear instructions so I can reproduce the run on my machine—package versions, command-line calls, and any preprocessing steps.
If you have previous experience combining Reinforcement learning with clustering or handling purely categorical data through embedding tricks, that will be a plus. Source files and reproducibility are the acceptance criteria; once I can retrain and obtain comparable clustering scores, the job is complete.