Loss Functions

Adding a LOSSES section to the config allows the user to select the loss function to be used during training. Multiple loss functions can be used by adding multiple elements to the list. You can set the weight of each loss function in WEIGHT_PER_LOSS. The default loss function is MSE.

DILATE

LOSSES:
  NAMES: ["DILATE"]
  ARGS: [{"alpha": 0.5, "gamma": 0.001}]

Reference

Shape and Time Distortion Loss for Training Deep Time Series Forecasting Models

Mean Absolute Error (MAE)

LOSSES:
  NAMES: ["MAE"]
  ARGS: [{}]

Mean Absolute Percentage Error (MAPE)

LOSSES:
  NAMES: ["MAPE"]
  ARGS: [{}]

Mean Squared Error (MSE)

LOSSES:
  NAMES: ["MSE"]
  ARGS: [{}]

Smooth MAE

LOSSES:
  NAMES: ["SmoothMAE"]
  ARGS: [{"beta": 0.11}]

Reference

Fast R-CNN