Languages: Python

Data Types:

_

Libraries:

_

Within this project I had to collect different sets of music data, comprising of MP3, WAV and MIDI files. I stored these files locally with relevant tagging that can be easily interpreted by the Neural Network I was going to feed this information through. First I needed to analyse the different formats to see what patterns could be identifiable and how they would be relevant to the reproduction of a new music sound file. Upon analysis the best dataset turned out to be with MIDI files, due to their easily interpretable nature, it made reproducing data possible. The other formats were still able to identify patterns, however, only white noise could be made when reproducing the data.

Creating a Convolutional Neural Network with the use of TensorFlow and Kera's, I was able to analyse a test and validation set of MIDI files to reproduce musical data. This data was very rudimentary and although it resembled that of the original data, there was still some kinks that could be worked out.

Overall this was a fun and interesting project, where I got to work with a range of different data collection and analysis techniques as well as ML.

Languages: Python

Data Types: Tetris Game-State

Libraries:

_

First we had to build the game up from scratch, so we were able to control the process more intently. For this we used PyGame, where one member of the team used the library to create the game pieces and game board. Implementing controls and a scoreboard, we were then ready to move onto the actual building of an agent that can learn to play the game. This was the basis to where we can create a depiction of the game state.

Moving on, we had a few meetings where we discussed the goals of the agent and how to measure its success when implementing a reinforcement learning algorithm. For this we needed to compile a list of rules in which the agent will need to follow and in what conditions the optimal choice can be made. For example, the agent should look to complete a row when it can, however, if it can place a block that can destroy 2 rows then that is better.

With this we then made the Neural Network in which these scenarios will be run and the best paths would be chosen. I headlined this part of the project, focusing on creating a model using PyTorch. I used a population size of 20, whereby the best 20% is chosen to move to the next generation. Using the rules we set in place, the agent started off slow but progress was made.

Ultimately, this was a fun project, and I enjoyed that the project was orientated around such an established game.

Languages: C++

Data Types:

_

Libraries:

_

I was given an Arduino in which I could collect information about the current state of the desk. I had a couple of solutions to this problem, both depending on the potential use of the project. The first was an ultra sonic sensor, this could sense when a presence is at the desk. The second was an RFID chip and sensor, this idea was shortlisted for a commercial use of the product, as each RFID chip could have different needs from the desk, which is perfect for hot desking. However, I decided against the RFID due to taking unnecessary additional effort from the user, as well as, memory concerns on the Arduino.

With the ultra sonic sensor, I needed to interpret the information it provides in a meaningful manner. As the sensor works by detecting how far away the nearest object is, I first needed to make sure the sensor is put in place with no obstacles in its way, then I needed to know the distance a person would be detected from in order ton trigger an event. For this I used a detection light, which displayed red when no one was present and green when there was. Within my code, I had a variable for the distance a person usually is from the sensor when they're present at the desk, with slack given 30cm further from that distance and 10cm closer. This allowed the device to comfortably recognise a presence at the desk.

The final part of the project was to get the device to interact with my Alexa and trigger a routine for the desk to follow. Upon research, there was a way to create a button in SmartThings that can be triggered by an access token in C++ Arduino code. In the code, I was able to trigger these buttons on and off. Using the Alexa application, I added these SmartThings buttons as smart devices. These buttons were then integrated into regular Alexa routines, like you would do for other smart home devices. Not only did this work, but I was able to trigger different routines based on different parameters. The additional routines I made for this project was based on the time of day, setting a different trigger for a routine suited accordingly based on whether it was night or day.

Overall, this was a fun little smart home product that could have the potential to be used in numerous ways. As mentioned it could be useful for hot desking multiple people and their different routines to a desk. For me it proves a useful tool when switching between my use of my desk when I needed to study during the day and wanted down time to play games in the evening.

Languages:

_

Data Types:

_

Libraries:

_

Using Android Studio I created a form for saving customer information as they check in for a club fitting, saving this data to an internal MySQL database.