Use AutoRun.py to run the FedPhoenix algorithm and the Baselines algorithm

Sure, I can help you with that. Here's an MD file that explains the usage of the provided script:

Experiment Configuration and Execution

Introduction

This script is used to configure and run various experiments for federated learning algorithms. It allows you to specify different datasets, models, algorithms, and various training-related parameters.

Usage

1. Dataset Related Parameters:

   • The datasets list specifies the datasets to be used for the experiments.
   • The dataset_classes dictionary defines the number of classes for each dataset.

2. Algorithm Related Parameters:

   • The algorithms list specifies the federated learning algorithms to be used.

3. Model Related Parameters:

   • The models list specifies the models to be used for the experiments.

4. Training Related Parameters:

   • lrs: Learning rates to be used.
   • epochs: Number of training epochs.
   • data_betas: Data distribution parameters, where 0.5 indicates IID data and other values indicate non-IID data.
   • weight_decays: Weight decay values.

5. Federated Learning Related Parameters:

   • num_users: Total number of clients.
   • frac: Fraction of clients participating in each round.

6. Specific Algorithm Parameters:

   • fp_convs: The total number of rounds for resetting in the Convolutional Layer of the FedPhoenix algorithm.Corresponding to r_s in the paper
   • resets: FedPhoenix algorithm's reset rate parameter.Corresponding to θ in the paper

7. Hardware Related Parameters:

   • gpus: List of GPU IDs to be used for the experiments.

8. Resource Management Parameters:

   • GPU_MEMORY_THRESHOLD: GPU memory usage threshold.
   • GPU_UTIL_THRESHOLD: GPU utilization threshold.
   • TASK_INIT_SLEEP: Task initialization wait time.
   • SCHEDULER_SLEEP_BUSY: Scheduler busy polling interval.
   • SCHEDULER_SLEEP_IDLE: Scheduler idle polling interval.

Parameter Validation

The script includes a validate_params() function that checks the validity of the parameter configuration, ensuring that the values are within the expected ranges.

Parameter Combinations

The script uses itertools.product() to generate all possible combinations of the specified parameters. It also dynamically generates the fixed_params dictionary based on the data_beta parameter, indicating whether the data is IID or non-IID.

GPU Manager

The script includes a GPUManager class that is responsible for monitoring and managing the status and task allocation of multiple GPUs. It provides methods to check GPU availability, get the best available GPU, and add/clean up running processes.

Main Function

The main() function is the entry point of the script. It initializes the GPUManager, schedules tasks on available GPUs, and manages the running tasks.