Reinforcement Learning/DQN expert to implement simple ridesharing/carpooling code (Python) – 4 days
Budget: $250 – $750 USD
1. A lot of the work has been done (at least 95%) in this Github repo (but you’ll need to fix small bugs here and there):
https://github.com/marina-haliem/Dynamic-RideSharing-Pooling-Simulator
2. Implement the repo (or use another that achieves same purpose), ensuring multiple passengers can use a single vehicle when it’s efficient to do so. Focus on delivering a clear, usable output. eg showing vehicle 1 picks up passenger 1, then picks up passenger 2, then drops of passenger 2 and finally passenger 1...output should be in any good format eg csv, json, excel etc (or could even be inserted into a table).
3. The project is about ridesharing (multiple passengers using single vehicle when feasible), not about implementing the Github repo….if you’ve done a similar previous projects and want to replace parts of the code, feel free to do so as long as you use RL/DQN.
4. Do some training with supplied NYC taxi data in the repo, generate a model, save the model and demonstrate it can take a simple input trips request file and produce an output (trip plan).
5. Ensure your solution can do the following (again, all already done in the repo):
- Move empty vehicles to areas of high demand
- Group passengers into vehicles balancing between profit and customer satisfaction (i.e. customers are happy to pay less for sharing but have threshold for how much extra times they'll spend in the vehicle due to sharing)
- vehicles already with passengers, when possible, should be able to efficiently insert a new passenger only minimally increasing the "in vehicle" time of other passengers
Finally, please, see attached "Requirements v3.docx" for clearer explanation.
https://github.com/marina-haliem/Dynamic-RideSharing-Pooling-Simulator
2. Implement the repo (or use another that achieves same purpose), ensuring multiple passengers can use a single vehicle when it’s efficient to do so. Focus on delivering a clear, usable output. eg showing vehicle 1 picks up passenger 1, then picks up passenger 2, then drops of passenger 2 and finally passenger 1...output should be in any good format eg csv, json, excel etc (or could even be inserted into a table).
3. The project is about ridesharing (multiple passengers using single vehicle when feasible), not about implementing the Github repo….if you’ve done a similar previous projects and want to replace parts of the code, feel free to do so as long as you use RL/DQN.
4. Do some training with supplied NYC taxi data in the repo, generate a model, save the model and demonstrate it can take a simple input trips request file and produce an output (trip plan).
5. Ensure your solution can do the following (again, all already done in the repo):
- Move empty vehicles to areas of high demand
- Group passengers into vehicles balancing between profit and customer satisfaction (i.e. customers are happy to pay less for sharing but have threshold for how much extra times they'll spend in the vehicle due to sharing)
- vehicles already with passengers, when possible, should be able to efficiently insert a new passenger only minimally increasing the "in vehicle" time of other passengers
Finally, please, see attached "Requirements v3.docx" for clearer explanation.