Configure Your Optimizer

We use the PyTorch optimizer object to train models within the framework of PyTorch and Lightning. As continual learning involves multiple tasks, each task is supposed to be given a optimizer for training. We could either use a uniform optimizer across all tasks or assign a distinct optimizer to each task.

Note

Optimization-based approaches in continual learning focus on designing mechanisms that adapt through optimization techniques. Typically, these strategies might involve using different optimizers for various tasks. However, the evolution of optimization can be integrated directly into the CL algorithm itself and we don’t particularly design β€œCL optimizers”.

Uniform Optimizer For All Tasks

To configure the uniform optimizer for all tasks for your experiment, link the optimizer field in the main YAML file to sub-config file in the sub-directory /optimizer/, and specify the class and its arguments in the sub-config. In this package, we just use the PyTorch built-in optimizers. They are useful enough that most of time we don’t need to implement ours, but just in case you do have the special need to custom your own torch.optim.Optimizer class, you can also implement and specify it for your experiment.

Here is an example:

./clarena/example_configs
β”œβ”€β”€ __init__.py
β”œβ”€β”€ entrance.yaml
β”œβ”€β”€ experiment
β”‚   β”œβ”€β”€ example.yaml
β”‚   └── ...
β”œβ”€β”€ optimizer
β”‚   └── sgd.yaml
...
example_configs/experiment/example.yaml
defaults:
  ...
  - /optimizer: sgd.yaml
  ...
example_configs/optimizer/sgd.yaml
_target_: torch.optim.SGD
_partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
lr: 0.01
weight_decay: 0.0

Make sure you include _partial_=True as the PyTorch optimizer need the model parameter as an argument to be fully instantiated and we now in the phase of configuration certainly don’t have that argument, the optimizer can be only partially instantiated.

Distinct Optimizer For Each Task

To configure distinct optimizer for each task for your experiment, the optimizer sub-config file should be a list of optimizer classes. Each class is assigned to a task and the length of the list should be the number of tasks.

example_configs/optimizer/sgd-10-tasks.yaml
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0
- _target_: torch.optim.SGD
  _partial_: True # partially instantiate optimizer without 'params' argument. Make sure this is included in any case!
  lr: 0.01
  weight_decay: 0.0

Supported Optimizers

We fully supported all optimizers defined in PyTorch. Please refer to PyTorch documentation to see the full list. To select a optimizer, replace with the class name. Please refer to PyTorch documentation of each class to know what fields (arguments) are required for the class.

Back to top