ysautoml.data.fyi

funtional

ysautoml.data.fyi.run_dsa

ysautoml.data.fyi.run_dsa(**kwargs)

Run dataset condensation using Differentiable Siamese Augmentation (DSA).

Parameters

• dataset (str): Dataset name (e.g., 'CIFAR10', 'CIFAR100').

• model (str): Backbone model (e.g., 'ConvNet').

• ipc (int): Images per class.

• eval_mode (str): Evaluation mode. Options:

  • 'S': same as training model

  • 'M': multi architectures

  • 'W': net width

  • 'D': net depth

  • 'A': activation function

  • 'P': pooling layer

  • 'N': normalization layer

• num_exp (int): Number of experiments.

• num_eval (int): Number of evaluation models.

• epoch_eval_train (int): Epochs for evaluation training.

• Iteration (int, default=1000): Training iterations.

• lr_img (float): Learning rate for synthetic images.

• lr_net (float): Learning rate for network parameters.

• batch_real (int): Batch size for real data.

• batch_train (int): Batch size for synthetic training.

• init (str): Initialization mode. Options: 'noise', 'real'.

• dsa_strategy (str): Augmentation strategy (comma-separated).

• data_path (str): Path to dataset root.

• device (str): Device ID, e.g. '0'.

• run_name (str): Experiment name.

• run_tags (str, optional): Tags for logging.

Returns

• dict: Containing results of all experiments:

  • save_path (str): Directory of logs and checkpoints.

  • accs_all_exps (dict): Recorded accuracies for each evaluation model.

  • eval_pool (list): Evaluation model pool.

  • num_exp (int): Number of experiments.

Examples

from ysautoml.data.fyi import run_dsa

run_dsa(
    dataset="CIFAR10",
    model="ConvNet",
    ipc=10,
    dsa_strategy="color_crop_cutout_flip_scale_rotate",
    init="real", lr_img=1.0, num_exp=5, num_eval=5,
    run_name="DSAFYI", run_tags="CIFAR10_10IPC", device="0", eval_mode="M",
)

---

ysautoml.data.fyi.run_dm

ysautoml.data.fyi.run_dm(**kwargs)

Run dataset condensation using Distribution Matching (DM).

Parameters

• (same as run_dsa, except default Iteration=20000)

Returns

• dict: Same format as run_dsa.

Examples

from ysautoml.data.fyi import run_dm

run_dm(
    dataset="CIFAR10",
    model="ConvNet",
    ipc=10,
    dsa_strategy="color_crop_cutout_flip_scale_rotate",
    init="real", lr_img=1.0, num_exp=5, num_eval=5,
    run_name="DMFYI", run_tags="CIFAR10_10IPC", device="1", eval_mode="M",
)