Implement paper: "MOVI: A Model-Free Approach to Dynamic Fleet Management"

[sash-ko]

MOVI, a Deep Q-network (DQN)-based framework that directly learns the optimal vehicle dispatch policy

MOVI: A Model-Free Approach to Dynamic Fleet Management

[yabirgb]

Is anyone working on this? I have a week free before the semester starts

[sash-ko]

I already started working on it and going to use this repo for all machine learning related papers https://github.com/sash-ko/ai-transportation

[sash-ko]

@yabirgb there is an interesting part is this paper - demand prediction with CNNs, it could be a topic on its own. A bit more details you can find in the technical report https://www.dropbox.com/s/ujqova12lnklgn5/dynamic-fleet-management-TR.pdf?dl=0

[yabirgb]

@yabirgb there is an interesting part is this paper - demand prediction with CNNs, it could be a topic on its own. A bit more details you can find in the technical report https://www.dropbox.com/s/ujqova12lnklgn5/dynamic-fleet-management-TR.pdf?dl=0

Ok I'm going to take a look, surely seems something interesting. Yesterday I finished the paper about T-share and I believe it has some ideas that can be mixed or used in the MOVI design although I need to reread it.

[yabirgb]

@sash-ko I'm playing around the topic of demand prediction using images and found this papers that might be interesting https://arxiv.org/pdf/1911.03441.pdf and https://arxiv.org/abs/1701.04245 (Yesterday I was trying to learn pytorch that is new for me)

[sash-ko]

In MOVI paper there is only a small chapter about CNNs for demand prediction

[yabirgb]

In MOVI paper there is only a small chapter about CNNs for demand prediction

I implemented the demand prediction as the paper mentions plus other variations that I made. Currently I'm training and comparing the models using the data from https://www1.nyc.gov/site/tlc/about/tlc-trip-record-data.page converting PULocation and DOLocation to a coordinate inside the region. I'll prepare a document with the comparison.

Moreover I'm implementing the model from https://arxiv.org/pdf/1911.03441.pdf to compare it too