src.environment.problem.SOO.NE.ne_dataset

Problem Difficulty Classification

Difficulty Mode

Training Set

Testing Set

easy

Deep networks (depth > 2)

Shallow networks (depth ≤ 2)

difficult

Shallow networks (depth ≤ 2)

Deep networks (depth > 2)

Note: Total 66 networks available. When difficulty is ‘all’, both sets contain all networks.

Module Contents

Classes

NE_Dataset

Introduction

This problem set is based on the neuroevolution interfaces in EvoX. The goal is to optimize the parameters of neural network-based RL agents for a series of Robotic Control tasks. We pre-define 11 control tasks (e.g., swimmer, ant, walker2D etc.), and 6 MLP structures with 0~5 hidden layers. The combinations of task & network structure result in 66 problem instances, which feature extremely high-dimensional problems (>=1000D).

API

class src.environment.problem.SOO.NE.ne_dataset.NE_Dataset(data, batch_size=1)[source]

Bases: torch.utils.data.Dataset

Introduction

This problem set is based on the neuroevolution interfaces in EvoX. The goal is to optimize the parameters of neural network-based RL agents for a series of Robotic Control tasks. We pre-define 11 control tasks (e.g., swimmer, ant, walker2D etc.), and 6 MLP structures with 0~5 hidden layers. The combinations of task & network structure result in 66 problem instances, which feature extremely high-dimensional problems (>=1000D).

Original paper

“EvoX: A distributed GPU-accelerated framework for scalable evolutionary computation.” IEEE Transactions on Evolutionary Computation (2024).

Official Implementation

NE

License

None

Initialization

Initializes the dataset object for single-objective optimization (SOO) problems.

Args:

  • data (list): A list of data items, where each item is expected to have a dim attribute.

  • batch_size (int, optional): The number of samples per batch. Defaults to 1.

Built-in Attributes:

  • data (list): Stores the input data.

  • batch_size (int): Stores the batch size.Defaults to 1.

  • N (int): The total number of data items.

  • ptr (list): List of starting indices for each batch.

  • index (np.ndarray): Array of indices for the data items.

  • maxdim (int): The maximum dimension found among all data items.Defaults to 0.

Notes:

Iterates through the data to determine the maximum dimension (maxdim) among all items.

static get_datasets(train_batch_size=1, test_batch_size=1, difficulty='easy', user_train_list=None, user_test_list=None, instance_seed=3849)[source]

Introduction

Generates training and testing datasets for the NE_Problem environment based on specified difficulty or user-provided lists.

Args:

  • train_batch_size (int, optional): Batch size for the training dataset. Defaults to 1.

  • test_batch_size (int, optional): Batch size for the testing dataset. Defaults to 1.

  • difficulty (str, optional): Difficulty level of the datasets to generate. Must be one of [‘all’, ‘easy’, ‘difficult’]. Defaults to ‘easy’.

  • user_train_list (list of str, optional): List of environment-depth identifiers to include in the training set. If provided, overrides difficulty for training set selection.

  • user_test_list (list of str, optional): List of environment-depth identifiers to include in the testing set. If provided, overrides difficulty for testing set selection.

  • instance_seed (int, optional): Random seed for instance generation. Defaults to 3849.

Returns:

  • NE_Dataset: The training dataset.

  • NE_Dataset: The testing dataset.

Raises:

  • AssertionError: If difficulty is not one of [‘all’, ‘easy’, ‘difficult’].

__getitem__(item)[source]

Introduction

Retrieves a batch of data samples corresponding to the given index.

Args:

  • item (int): The index of the batch to retrieve.

Built-in Attribute:

  • self.ptr (list or array-like): Maps batch indices to starting positions in the dataset.

  • self.index (list or array-like): Contains indices of the data samples.

  • self.batch_size (int): The number of samples in each batch.

  • self.N (int): The total number of data samples.

  • self.data (list or array-like): The dataset from which samples are retrieved.

Returns:

  • list: A list containing the data samples for the specified batch.

Raises:

  • IndexError: If item is out of range of available batches.

__len__()[source]

Introduction

Returns the number of elements in the dataset.

Returns:

  • int: The total number of elements in the dataset.

__add__(other: src.environment.problem.SOO.NE.ne_dataset.NE_Dataset)[source]

Introduction

Combines two NE_Dataset instances by adding their data attributes and returns a new NE_Dataset with the same batch size.

Args:

  • other (NE_Dataset): Another NE_Dataset instance to be added.

Returns:

  • NE_Dataset: A new dataset containing the combined data of both instances.

Raises:

  • AttributeError: If other does not have a data attribute.

shuffle()[source]

Introduction

Randomly shuffles the indices of the dataset, updating the internal index array.

Built-in Attribute:

  • self.N (int): The number of elements in the dataset.

  • self.index (np.ndarray): The array storing the current order of indices.

Returns:

  • None

Notes:

This method uses np.random.permutation to generate a new random ordering of indices for the dataset.