We are ending near the end of the term, or in other words, the development phase. While this ten-week, two sprint phase was considerably productive, our team on the top-n music genre classification neural network project still has lots to get done.
At the beginning of the term, our team created a development plan to set our goals for our ten-week span. The plan was straightforward. “The alpha version of the Top-n Neural Network must be able to classify a single
audio clip by genre as accurately as possible.” Our project could be seen as a similarly to a mathematical function, which takes an input, and give an output. The input for our project is simply a song followed by the output of its genre.
Not included in the alpha are any of the ‘UI’ or ease of use functions – no
directory scanning, no ‘only this many genres’ CLI arguments, and the like. This does not represent a deviation from the Design Document, as these features will be added in the Beta release.
Our progress as of now has been taken over by a lot of working with the inner neural network. We still have bugs appearing within model, which have actually become a more recent problem with the addition of different edge cases we try to factor in. On the other hand, these ‘setbacks’ our neural network team, which includes largely of myself, does not affect those of the UI team. The UI team has in fact already begun with the command line interface for the project.
By far, the most challenging part of this project has been learning the ways of building the neural network model with little to no experience of building models. Understanding the different aspects of the model and how they are portrayed in the results has taken a lot of time, reading, and research. Perhaps the most time consuming is all the testing, and this becomes even more difficult when a lot of our testing takes so much time, with all the large amounts of data we have to work with.
