Forecasters

class forecasters.Forecaster(model: <sphinx.ext.autodoc.importer._MockObject object at 0x7fec74944a90>, loss=<sphinx.ext.autodoc.importer._MockObject object>, optimizer=<sphinx.ext.autodoc.importer._MockObject object>, n_epochs=1, device='cpu', checkpoint_path='./', verbose=True, nan_budget=5)

Handles training of a PyTorch model. Can be used to generate samples from approximate posterior predictive distribution.

Parameters:

• model – Deep4cast neural network of class models
• loss – PyTorch distribution
• optimizer – PyTorch optimizer
• n_epochs – number of training epochs
• device – device used for training (cpu or cuda)
• checkpoint_path – path for writing model checkpoints
• verbose – switch to toggle verbosity of forecaster during fitting
• nan_budget – how many time the forecaster will try to continue batch training when NaN encountered.

embed(dataloader, n_samples=100)

Generate embedding vectors.

fit(dataloader_train, dataloader_val=None, eval_model=False)

Fit model to data.

predict(dataloader, n_samples=100)

Generate predictions.