AutoClipper

class lib.model.autoclip.AutoClipper(clip_percentile: int, history_size: int = 10000)

Bases: object

AutoClip: Adaptive Gradient Clipping for Source Separation Networks

Parameters:

  • clip_percentile (int) – The percentile to clip the gradients at

  • history_size (int) – The number of iterations of data to use to calculate the norm Default: 10000

References

Adapted from: https://github.com/pseeth/autoclip original paper: https://arxiv.org/abs/2007.14469

Methods Summary

__call__(parameters, *args)

Call the AutoClip function.

Methods Documentation

__call__(parameters: list[Parameter], *args) → None

Call the AutoClip function.

Parameters:

  • parameters (list[Parameter]) – The parameters to clip

  • args – Unused but for compatibility

Return type:

None

__call__(parameters: list[Parameter], *args) → None

Call the AutoClip function.

Parameters:

  • parameters (list[Parameter]) – The parameters to clip

  • args – Unused but for compatibility

Return type:

None