Loss functions
Numpy-based loss functans that can be used by environments or non-pytorch models.
pinball_loss
pinball_loss (Y_true:numpy.ndarray, Y_pred:numpy.ndarray, underage_cost:ddopai.utils.Parameter|numpy.ndarray, overage_cost:ddopai.utils.Parameter|numpy.ndarray)
Pinball loss calculating the cost of underestimating and overestimating the target value based on specific underage and overage costs. Used to evaulate the Newsvendor cost.
| Type | Details | |
|---|---|---|
| Y_true | ndarray | |
| Y_pred | ndarray | |
| underage_cost | ddopai.utils.Parameter | numpy.ndarray | |
| overage_cost | ddopai.utils.Parameter | numpy.ndarray | |
| Returns | ndarray | returns the cost per observation |
quantile_loss
quantile_loss (Y_true:numpy.ndarray, Y_pred:numpy.ndarray, quantile:Union[float,ddopai.utils.Parameter])
Similar evaluation function to the pinball loss, but with the quantile of range [0, 1] as a parameter instead of SKU-specific cost levels for underage and overage.
| Type | Details | |
|---|---|---|
| Y_true | ndarray | |
| Y_pred | ndarray | |
| quantile | Union | |
| Returns | ndarray | returns the cost per observation |