PeterSteiglechner / next-bar

A tool to help your group decide which bar/café to hit next incorporating on preferences-based decision and randomness.

next-bar

Code deployed at

https://www.pythonanywhere.com/user/petersteiglechner/files/home/petersteiglechner/mysite/

Website: http://petersteiglechner.pythonanywhere.com/

Description

The tool is designed to suggest a bar/café from a pre-defined set taking the users' preferences into account, but maintaining some degree of randomness to foster exploration.

The $N$ users rate all pre-selected $n$ bars. Ratings $r_i$ of a user $i$ are hard-coded and range from $1$ (least favourite bar) to $n$ (favourite bar) or from $0$ to $100$. They can be updated through the UI. The probability that a bar/café $x$ is chosen is

$$ p(x) = 1/R \cdot \sum_{i} , r_i(x) $$

where $R$ is the normalisation $\sum_{i} , r_i(x)$.

Issues and plans

UI/Backend: Flexible users

• create a new "User" ✔️
• store user ratings.
• enable user selection and load their stored ratings.

UI: Rating input

• make ratings adjustable ✔️

UI: Bar selection/exclusion

• create a tickbox behind each bar/café to enable manual selection or exclusion of certain bars/cafés.

UI: Bar-set selection

• enable to switch between "cafés" and "bars" with a simple click.
• enable to switch between "cafés" and "bars" with a simple click.

Backend: 0-10 rating

• Instead of ranking the bars from 1 to n, one could give scores from 0 to 10.

Backend: "no rating"

• enable the possibility to assign no rating to a bar $r_i = NA$. The probability needs to be adjusted somehow.

Backend: Decision-function

• slider from "secure choice" to "experimental choice" --> parameter $f$ in the range from 0 to a large value.

The new decision function is:

$$ p(x) = 1/R \cdot \sum_{i} , r_i(x)^{f} $$

For $f=0$, all bars are equally likely. For $f=1$, the average rating determines the probability. For $f=\infty$, the overall best-rated bar is chosen.

• Alternative: slider "min rating" from 1 to n --> parameter $r_{min}$. All bars with any rating $r_i < r_{min}$ are excluded (and the normalisation is adjusted accordingly).