Inspiration
It simulates people randomly moving around. Once there are too many people close together, an "illegal gathering" is called, and the simulation stops. The goal is to predict how long it will take for the simulation to stop, given the number of smaller gatherings of each size. People are considered to be in a gathering if they are close to a common point measured in the manhattan distance.
What it does
At each step of the simulation, every person moves one unit in a random direction, unless that position is taken or out of the grid. After every person moved, the number of gatherings of 0 people (=empty zones), 1 person, 2 people, up to 5 people are counted. If there is a gathering of at least 6 people, then the simulation is stopped. Otherwise, these 6 numbers are the number of remaining steps until the end of the simulation given as label, and fed to the ML algorithm. The goal of the ML algorithm is to see only these 6 values, and guess the label. We apply a variation of the k-Nearest Neighbours algorithm, in our case the parameter M is set to a default value of 10 and express the number of closest neighbours to find, we then estimate the label by calculating the median of the labels for the previously selected neighbours.
How we built it
Python and jupyter notebook.
Challenges we ran into
NN not predicting the problem well, we therefore switched to another algorithm.
Accomplishments that we're proud of
Results make sense and are close to what is expected.
What we learned
Practice with pytorch.
What's next for Simulation and Neural Networks
Implement a new algorithm.
Built With
- knn
- python
Log in or sign up for Devpost to join the conversation.