tsts.schedulers

Functions

class tsts.schedulers.build_scheduler[source]

Build learning rate scheduler.

Parameters

  • optimizer (Optimizer) – Target optimizer

  • cfg (CN) – Global config

Returns

Built learning rate scheduler

Return type

Scheduler

Classes

class tsts.schedulers.CosineAnnealing[source]

Cosine annealing scheduler implementation.

Example

SCHEDULER:
  NAME: "CosineAnnealing"
  T_MAX: 10
Parameters

  • optimizer (Optimizer) – Target optimizer

  • T_max (int) – Maximum number of iterations (from pytorch)

  • eta_min (float, optional) – Minimum learning rate (from pytorch), default 0.0

class tsts.schedulers.IdentityScheduler[source]

Dummy scheduler.

Example

SCHEDULER:
  NAME: "IdentityScheduler"
Parameters

optimizer (Optimizer) – Target optimizer

class tsts.schedulers.StepScheduler[source]

Step scheduler implementation.

Example

SCHEDULER:
  NAME: "StepScheduler"
  T_MAX: 10
Parameters

  • optimizer (Optimizer) – Target optimizer

  • T_max (int) – Maximum number of iterations (from pytorch)

  • eta_min (float, optional) – Minimum learning rate (from pytorch), default 0.0