Predicting the unpredictable stock market
Python, PyTorch, OpenAI Gym, Keras, Tensorflow, Scikit, Jupyter, Google Colab, Docker
October 2020: Attention is All You Need
Analyzing the state of the art paper about Transformers, I played around with the Tensorflow and PyTorch implementations. The results for language translation were apparent, but I tried to take it further and apply the model towards time series. While stock figures are commonly accepted as unpredictable due to countless factors and human emotions, I built on my previous LSTM exploration and achieved better predictions.
Apr 2020: StockChek
An exploration of Long Short Term Memory for time series forecast, notably within the stock market. Many iterations of neural networks with various parameters and setups were attempted. Moderate short-term portfolio gain was achieved when paired with market knowledge.