Shortcuts

MatrixScaler

class torch_uncertainty.post_processing.MatrixScaler(num_classes, model=None, init_w=1, init_b=0, lr=0.1, max_iter=200, device=None)[source]

Matrix scaling post-processing for calibrated probabilities.

Parameters:
  • model (nn.Module) – Model to calibrate.

  • num_classes (int) – Number of classes.

  • init_w (float, optional) – Initial value for the weights. Defaults to 1.

  • init_b (float, optional) – Initial value for the bias. Defaults to 0.

  • lr (float, optional) – Learning rate for the optimizer. Defaults to 0.1.

  • max_iter (int, optional) – Maximum number of iterations for the optimizer. Defaults to 100.

  • device (Optional[Literal["cpu", "cuda"]], optional) – Device to use for optimization. Defaults to None.

Reference:

Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. On calibration of modern neural networks. In ICML 2017.

fit(calibration_set, save_logits=False, progress=True)

Fit the temperature parameters to the calibration data.

Parameters:
  • calibration_set (Dataset) – Calibration dataset.

  • save_logits (bool, optional) – Whether to save the logits and labels. Defaults to False.

  • progress (bool, optional) – Whether to show a progress bar. Defaults to True.

set_temperature(val_w, val_b)[source]

Set the temperature to a fixed value.

Parameters:
  • val_w (float) – Weight temperature value.

  • val_b (float) – Bias temperature value.