Adds a dataloader so that data will only be loaded into memory when necessary. A few additional notes:
Normalizing values are now calculated in the engineer, since the full dataset is never loaded into memory
The linear regression now uses SGD (instead of ordinary least squares) since everything happens in minibatches
Functionality to make training and validation sets, for early stopping (with early stopping optionally implemented in the linear regression as an example)
Tested with unit tests, as well as by training the linear regression model using VHI, CHIRPS precip and GLEAM (E, SMroot and SMsurf) data, with and without early stopping
Resolves #38
Adds a dataloader so that data will only be loaded into memory when necessary. A few additional notes:
Tested with unit tests, as well as by training the linear regression model using VHI, CHIRPS precip and GLEAM (
E,SMrootandSMsurf) data, with and without early stopping