livreq / dynamiccml

DynamicCML

This code repo contains the implementation of the DCML algorithm presented in the NeurIPS 2023 paper:

On the Stability-Plasticity Dilemma in Continual Meta-Learning: Theory and Algorithm by Qi Chen, Changjian Shui, Ligong Han, and Mario Marchand.

Prerequisites

To run the code, make sure that the requirements are installed.

pip install -r requirements.txt

The code was written to run in Python 3.6 or in a more recent version.

Code Structure

.
|-- config
|   `-- yamls
|       |-- omniglot
|       `-- synbols               
|-- output
|   `-- models
|       |-- omniglot_maml.model
|       `-- synbols_maml.model
|-- run_guassian.sh
|-- run_osaka.sh
`-- src
    |-- dataloader
    |   |-- osaka_datasets.py
    |   `-- toy_datasets.py
    |-- envdetector
    |   |-- bocd.py
    |   `--env_change_detection.py
    |-- learners
    |   |-- dmogd.py
    |   |-- __init__.py
    |   |-- maml.py
    |   |-- model.py
    |   `-- osaka_model.py
    |-- main
    |   |-- args.py
    |   |-- run_gaussian.py
    |   `-- run_osaka_benchmark.py
    `-- utils

Continual Meta-Learning Framework

Experimental Setting

Run Experiment

Moving 2D Gaussian

    python src/main/run_gaussian.py --sample_num 100 --K 2 --hazard 0.1 --method 1 --seed 168

or

   sh run_gaussian.sh

OSAKA

1. Create your wandb account, a new project, and a wandb key
2. Run the code w.r.t different algorithms ("DCML_oracle DCML_win CMAML_pre_kwto_acc online_sgd fine_tuning MetaCOG MetaBGD BGD") For example:

sh run_osaka.sh "DCML_oracle" 0.0 0.2 "synbols" wandb_project_name  wandb_key

Reference

Acknowledgements

The code for testing the OSAKA benchmark was adapted from https://github.com/ServiceNow/osaka/tree/camera_ready.

dynamiccml's Issues

Error in dmogd.py Line 620: Mismatch in Number of Returned Values from self.SGD

Hello,

I have noticed an issue in your dmogd.py file at line 620:

w_t, train_per, test_per, train_loss, test_loss = self.SGD(x_train, y_train, x_test, y_test, test)

In the self.SGD method, only four values are returned, but you are trying to unpack five values here. The variable w_t is missing in the returned values from the self.SGD method. This appears to be an error.

Could you please update the code to resolve this issue?

Thank you!