Forecasting with time series:
-
Predicting Google stock in 2017 (straight forward regression model) First attempt of an RNN project, the data was not extremely easy to predict, but it was reasonable. The main concepts implemented are preprocessing with pandas and sklearn as well as learning with RNN's that contain LSTM cells and dropout layers.
-
Predicting Sunspots (very complicated regression problem)
- Full project with hours of experimentation that combined most deep learning topics. Non-ML topics implemented included but were not limited to preprocessing series data from csv file, windowed dataset, plotting series data and loss functions. ML topics implemented were RNNs, Convolutional Layers, multi-LSTM layers, bidirectionality of LSTMs, lambda layers, callback functions for learning rate optimization, and hyper parameter tweaking.
- This was potentially one of the most time consuming deep learning projects that I had to tackle because the loss would not be easily reduced and I had to utilize everything I learned to make a respectable model.
As always, thanks for taking an interest in my learning process! I challenge you to improve my final sunspots model!