Quick Start

Get up and running with AutoMAX in three steps.

Step 1 — Install

conda create -n AutoMAX python=3.10
conda activate AutoMAX
conda install gxx_linux-64 gcc_linux-64 swig
pip install -r requirements.txt

Step 2 — Write a config

Create a config.yaml file:

dataset:
  name: cifar10
  eval_splits: [val, test]
  kwargs:
    imratio: 0.1

model:
  name: resnet18
  num_classes: 1

metrics:
  - AUROC

training:
  experiment_name: my_first_run
  loss: AUCMLoss
  optimizer: PESG

automax:
  name: my_first_run
  n_trials: 5

Step 3 — Run AutoMAX

python -m src.auto_trainer --config_file config.yaml

You can override any training field directly on the command line:

python -m src.auto_trainer --config_file config.yaml --epochs 50 --seed 0

Entry Points

Command	Use case
`src.auto_trainer`	AutoTune on standard image datasets (MedMNIST, CIFAR, CheXpert, …)
`src.auto_transformers_trainer`	AutoTune with Transformer backbones (e.g. RIP-Dataset)