Currently, retailers lack the possibility to test, develop, and evaluate their algorithms appropriately before releasing them into the real world. At the same time, it is challenging for researchers to investigate how pricing strategies interact with each other under heavy competition.
We built an open platform to simulate dynamic pricing competition allowing both practitioners and researchers to study the effects of automated repricing mechanisms competing with each other using market scenarios that mimic real-world marketplaces.
We built the platform in a way that one can participate and deploy own merchants with only a few lines of Python code. It allows merchants to deploy the full width of pricing strategies, from simple rule-based strategies to more sophisticated data-driven strategies using machine learning. For practitioners, the platform further provides a possibility to evaluate their pricing strategies appropriately before releasing them in production.
For more information about the platform and publications, see the our chair's project site on Dynamic Pricing under Competition.
On the master branch, this repository contains the docker-setup that allows running the simulation locally. On the gh-pages branch one can find the API specification of all components.
Repositories
- Management UI: https://github.com/hpi-epic/pricewars-mgmt-ui
- Consumer: https://github.com/hpi-epic/pricewars-consumer
- Producer: https://github.com/hpi-epic/pricewars-producer
- Marketplace: https://github.com/hpi-epic/pricewars-marketplace
- Merchant: https://github.com/hpi-epic/pricewars-merchant
- Kafka RESTful API: https://github.com/hpi-epic/pricewars-kafka-rest
- Analytics: https://github.com/hpi-epic/pricewars-analytics
- Docker
- If you are on Linux read this for running docker as non-root user.
- Docker Compose
Clone the repository and its subrepositories:
git clone --recursive git@github.com:hpi-epic/pricewars.git
For the next step bring a fast internet line and some time. This can take up to 30 minutes at the first-time setup. Build docker images and containers with the following command.
cd pricewars
docker-compose build
The Price Wars platform can be started with:
docker-compose up
This will start all services and one example merchant.
You can shut down the platform with CTRL + C or docker-compose stop.
Warning: There might be routing problems if the docker network overlaps with your local network.
If this is the case, change the ip address in docker-compose.yml under the networks entry.
After starting the Pricewars platform with docker-compose up, it can be controlled with the Management UI
- [Optional] Configure available products in the Config/Producer section
- Start the Consumer
- Merchants are trading products now. The Dashboard shows graphs about sales, profits and more.
In the Merchant repository you can learn how to build your own merchant and run it on the platform.
Run the following commands to run the platform in a clean state.
docker-compose down
docker-compose up
First, stop your running containers.
docker-compose stop
Update repositories.
git pull
git submodule update
Rebuild all images that have changed:
docker-compose build
You can see the status of the containers with docker-compose ps.
In case a container is not running, you can see its last logs with docker-compose logs <service name>.
- Postgres: Bring the platform in a clean state with
docker-compose downand run it again. - Zookeeper / Kafka: If you just stopped some older containers: Wait! There is a timeout for Zookeeper to notice that Kafka has been stopped (timestamp-based, so it works even if Zookeeper is not running). Bring the platform in a clean state with
docker-compose rm --stopand run it again. - Others: Try to read the logs or read on.
- Reset the containers and the network:
docker system prune(and restart the Docker service). - Terminate Docker and ensure, that all docker processes are stopped (especially the network service).
- Restart your computer and wait (even though it might be slow) for about five to ten minutes.
- Reset Docker to factory defaults (should be your last attempt, as this requires re-building of all images):
- macOS: Click on "Preferences" > "Reset" > "Reset to factory defaults"
For details regarding the deployment of the component, we kindly refer to the deployment section of the microservice specific README.md file. The links can be found above.
You can run a benchmark on the Price Wars platform with the benchmark tool benchmark.py. This tool allows to run the platform in a given configuration for a specific time period. Afterwards, results of this run are written to the output directory.
Firstly, install necessary Python libraries:
python3 -m pip install -r helper_scripts/requirements.txt
Example command:
python3 helper_scripts/benchmark.py --duration 30 --output <output directory> --merchants <merchant A command> <merchant B command>
This starts the whole platform and two merchants to compete against each other for 30 minutes.
As merchant start command you can use for example: "python3 merchant/merchant.py --strategy cheapest --port 5000".
The quotes must be included.
Run python3 helper_scripts/benchmark.py --help to see all arguments.
There are different ways to develop the containerized marketplace services. These approaches have different trade-offs between setup time and time to run the platform with the new code.
This method does not require any extra setup. The workflow is to modify code, then rebuild the docker images. Use the following commands:
docker-compose down
docker rmi <name_of_image>
On the next start of the platform, the new image will be built.
Rebuilding an image takes a few seconds up to over a minute depending on the service.
If rebuilding the image to too tedious, an alternative approach is mounting the executable (or source code) onto the docker container. This way, the program can be built locally and is used by the container on the next start.
The directory can be mounted in the docker-compose.yml file with the volumes option.
It is possible to run a service natively. This removes the docker abstraction and is the fastest way to restart the service with a new version.
However, the service must be reconfigured to connect to the docker services and vice versa.
Visit the corresponding subrepository on how to run a service natively and how to configure it.