An unusual journey learning about NNs for a PhD thesis
https://github.com/DidierRLopes/UnivariateTimeSeriesForecast
On 14th February of 2019, my previous Maths and Statistics teacher — Filipe -sent me a messaged because of a Linkedin post I shared about work I was doing in python.
It turns out that Filipe was looking for someone to help him with his PhD thesis, in specific, with the programming side of it. The challenge was to study diverse models (from classical to neural networks) and assess their forecasting performance. Since time series prediction was always a topic that I found fascinating and hadn’t had time to study, I thought this would be the perfect timing to do so.
So from February 2019 onwards, this exciting journey started. I was working full-time so in order to be able to take part in this, I was only sleeping 4/5h a day. I started reading a lot of books and practicing my python coding skills in order to be more helpful. Then around June, we started working together on the code. We had around 2–3h discussions a couple times a week where we would discuss the point of the situation code-wise and where we wanted to be, we kept in touch about this every day.
From the repo, which is open source here, you can see that we explored: Exploratory Data Analysis; ARIMA and SARIMA; Exponential Smoothing; Deep Neural Network…
