We're two math engineering students who really wanted to get into neural networks this weekend. We also thought it would be cool to jump into something science project style and without knowing how it would turn out. For those reasons, we wanted to see if we could train a neural network to guess the temperature of a geographical location based on temperature data of a few surrounding points. This would have the potential to predict weather in areas that don't have a permanent or no longer have a weather station.
What it does
It takes in temperature and precipitation data from four neighbouring weather stations as training data. The software then learns how to weight data appropriately and we compare it's guesses on testing data (data from the same four station over a different time period where the program never sees the fourth "to be guessed" location's data).
How we built it
We built it in python with no external libraries for the neural network except for numpy (we really wanted to do the math ourselves).
Challenges we ran into
Neither of us are python programmers so there was a lot of learning the basics this weekend.
Accomplishments that we're proud of
The fact it worked it all. We have never worked with neural networks before or python so it was really cool to see it all come together.
What we learned
Basics of python and some really cool math.
What's next for Down to Neural Netword (DNN)
Improving the accuracy (through the literal optimizer function in the network) and getting it to work with non-scalar output systems. This also might inspire our fourth year thesis project for next year!