Segmentations

class adelecv.api.optimize.segmentations.HPOptimizer(hyper_params, num_trials, device, dataset)

Class for hyperparams search and model training

Parameters:

• hyper_params (HyperParamsSegmentation) – Dataclass with hyperparams of models

• num_trials (int) – Number of iterations algorithm (the number of models with pruned models)

• device (str) – GPU or CPU

• dataset (SegmentationDataset) – Created dataset

optimize()

Run optimize.

Return type:

None

property stats_models: pandas.DataFrame

Get stats models after training.

Returns:

Info about trained models

adelecv.api.optimize.segmentations.get_hp_optimizers()

Get list of names optimizer strategy

Returns:

List of names hyperparams optimizer strategy

Return type:

list[str]

class adelecv.api.optimize.segmentations.HyperParamsSegmentation(*args, **kwargs)

Dataclass with set of hyperparams. For all possible values see Segmentation.

Parameters:

• strategy – Name optimizer strategy (optuna name Sampler).

• architectures – List of name model (Unet, DeepLabV3, e.g.)

• encoders – List of name encoders (resnet18, mobilenet, e.g.)

• pretrained_weights – List of pretrained weights (imagenet or None), only str! None=’None’

• loss_fns – List of names loss func (DiceLoss, JaccardLoss, e.g.)

• optimizers – List of names pytorch optimizers (AdamW, Adadelta, e.g.)

• epoch_range – range epoch (from - to)

• lr_range – range lr (from - to)

• optimize_score – Score for optimizing optimizers (optuna score)