hangulu / onthetown

On The Town: a web application designed to facilitate event planning between friends.

On The Town: Social Planning As A Search Problem

On The Town (OTT) is a web application that allows people in the same city, yet in different locations, to meet up in a location that is relatively equidistant from all of the people, and also satisfies users' preferences for type of activity, budget, and rating.

Given a set of users spread out in various locations and information on venues and activities in a desired destinations, OTT optimizes the process of meeting up with those that you care about. The problem of social planning here is well encoded as a search problem. The developers of OTT chose from a variety of search algorithms (including depth-first search, breadth-first search, uniform cost search, greedy search, and A* search) to output a list of 7 places that not only satisfy users' preferences, but are also diverse in their offerings. Thus, OTT gives users a useful and varied set of options to execute.

The production version of OTT found here utilizes Greedy Search. All algorithms can be found in algorithms.py.

The following is an example of the expected behavior: imagine 6 friends scattered around New York City: uptown, downtown, midtown, etc. Based on the factors they decide, both through polls and preferences within their user profiles, OTT minimizes the distance that each will have to travel in order to satisfy their desires, while also satisfying other preferences and the goal of diversity. Given all of the users’ preferences, OTT outputs 7 locations. Then, the friends meet up and have a fun time.

The core features of this application rely on the Google Places API, and search algorithms. The stack is as follows:

• Backend: Python, Flask, SQL
• Frontend: HTML, JavaScript, CSS, Jinja2

---

Installation

OTT runs on Python 2.7, and requires the following packages:

• flask
• flask-session
• requests
• werkzeug
• requests-cache
• redis
• numpy
• scipy
• matplotlib
• pandas
• plotly

There are two ways to install these dependencies:

1. Pipenv [recommended]. Install Pipenv with brew install Pipenv if you're using Homebrew on MacOS or Linuxbrew on Linux, or sudo dnf install pipenv if you're using Fedora28. Otherwise, navigate here for more instructions. Next, navigate to the root directory of this repository and run pipenv install to install all dependencies at once and create a virtual environment for the project.
2. pip. Download pip from with the following command in a Terminal window: curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py. Next, run python get-pip.py. Finally, navigate to the root directory of this repository and run pip install -r requirements.txt to install all dependencies at once.

To initialize the database to store users' information:

1. Ensure that you have deleted users.db in the server/ directory.
2. Run db_init.py
3. Run redis-server

To be able to use the Google Places API and thus OTT, you must create a file called config.py and place the following line in it:

api_key = "[YOUR GOOGLE PLACES API KEY]"

Replace [YOUR GOOGLE PLACES API KEY] with your Google Places API key, which can be found here. Ensure that the API key is put in quotation marks, and that the file is saved in the root directory of the repository.

Finally, to run the web application, run python app.py in a new Terminal window.

---

Algorithm Testing

testing.py includes tests of all the OTT algorithms. This file allows one to check if the parties and users are being instantiated correctly, to investigate the process of that data structure creation, and test each of the algorithms individually. To run this file, type python testing.py into a Terminal window.

While testing.py tests the basic functionality of the application, a Jupyter notebook file was included to perform robust simulations. evaluation.ipynb performs many simulations of each algorithm on random parties. This file was used to generate the data used in the Results section of the accompanying paper. The data generated can also be found in the file simulation_results.csv. While there is an aspect of randomness to the data generation process (parties of random properties are instantiated), the data analysis process can be replicated by running the \texttt{analysis.ipynb} file, which utilizes the data the developers generated.

As mentioned in the Results and Discussion sections, BFS takes a very long time to run, so there were time limits imposed on its evaluation. These time limits are fungible, and one may also opt to not test BFS by removing or commenting out the lines of code that respond to its evaluation. For an idea of how long these may take, the developers performed overnight trials to gather the included data.

---

Authors

Hakeem Angulu, Harvard College Class of 2020 [email protected]

Louie Ayre, Harvard College Class of 2020 [email protected]

Amadou Camara, Harvard College Class of 2020 [email protected]