nip.scenario_base.pretrained_models.PretrainedModel

class nip.scenario_base.pretrained_models.PretrainedModel(hyper_params: HyperParameters, settings: ExperimentSettings)

Base class for pretrained models, used to generate embeddings for datasets.

Pretrained models can by used to generate embeddings for datasets, which can then be used by agents instead of raw data. This means the agent starts with a good level of knowledge about the data.

Parameters:

• hyper_params (HyperParameters) – The parameters for the experiment

• settings (ExperimentSettings) – The settings for the experiment

• attributes (Class)

• ----------------

• name (str) – The name of the model, which should uniquely identify it

• dataset (str) – The name of the dataset the model was trained for

• allow_other_datasets (bool, default=False) – Whether the model can be used for datasets other than the one it was trained on

Methods Summary

__init__(hyper_params, settings)
generate_dataset_embeddings(datasets[, ...])	Load the model and generate embeddings for the datasets.

Attributes

allow_other_datasets
name
dataset

Methods

__init__(hyper_params: HyperParameters, settings: ExperimentSettings)

abstract generate_dataset_embeddings(datasets: Iterable[TensorDictDataset], delete_model: bool = True) → Tensor

Load the model and generate embeddings for the datasets.

Parameters:

• datasets (Iterable[TensorDictDataset]) – The datasets to generate embeddings for

• delete_model (bool, default=True) – Whether to delete the model after generating embeddings

Returns:

embeddings (torch.Tensor) – The embeddings for the datasets