schema
Configuration for the dataset schema. The dataset schema defines the content of a raw dataset, including its metadata for zarr storage. It does not define which subset of the data is used at runtime (see TaskSpaceConfig for that).
dataclass
¶
Configuration for the dataset schema.
Attributes:
| Name | Type | Description |
|---|---|---|
_target_ |
str
|
Import path instantiated by Hydra. |
zarr_path |
str
|
Path to save/load the zarr file. |
metadata |
DatasetMetadataConfig
|
Metadata of the raw dataset. |
dataset_type |
str
|
Type of dataset (e.g., 'libero', 'tso', 'metaworld'). |
dataclass
¶
CsvDatasetSchemaConfig(_target_=MISSING, zarr_path=MISSING, metadata=MISSING, dataset_type=MISSING, dataset_folders=MISSING)
Bases: DatasetSchemaConfig
Configuration for CSV dataset schema.
Attributes:
| Name | Type | Description |
|---|---|---|
dataset_folders |
list[str]
|
List of dataset folder paths. |
dataclass
¶
Hdf5DatasetSchemaConfig(_target_=MISSING, zarr_path=MISSING, metadata=MISSING, dataset_type=MISSING, hdf5_paths=MISSING)
Bases: DatasetSchemaConfig
Configuration for HDF5 dataset schema.
Attributes:
| Name | Type | Description |
|---|---|---|
hdf5_paths |
list[str]
|
List of paths to HDF5 files. |
dataclass
¶
SyntheticDatasetSchemaConfig(_target_=MISSING, zarr_path=MISSING, metadata=MISSING, dataset_type=MISSING, task_name=MISSING, num_episodes=1000, seed=42, image_size=64, num_modes=2, trajectory_length=60, noise_std=0.01, num_styles=1, mode_weights=None, num_rollouts=200)
Bases: DatasetSchemaConfig
Configuration for synthetic multimodal benchmark schema.
Attributes:
| Name | Type | Description |
|---|---|---|
task_name |
str
|
Synthetic task variant to generate. |
num_episodes |
int
|
Number of generated episodes. |
seed |
int
|
Random seed for reproducibility. |
image_size |
int
|
Generated image side length in pixels. |
num_modes |
int
|
Number of behavior modes in the trajectory distribution. |
trajectory_length |
int
|
Timesteps per generated trajectory. |
noise_std |
float
|
Standard deviation of the trajectory noise. |
num_styles |
int
|
Number of visual styles used for conditional variants. |
mode_weights |
list[float] | None
|
Sampling weight per behavior mode. |
num_rollouts |
int
|
Rollouts sampled per mode when evaluating coverage metrics. |
dataclass
¶
LeRobotDatasetSchemaConfig(_target_=MISSING, zarr_path=MISSING, metadata=MISSING, dataset_type=MISSING, dataset_path=MISSING)
Bases: DatasetSchemaConfig
Configuration for LeRobot dataset schema V30
Attributes:
| Name | Type | Description |
|---|---|---|
dataset_path |
str
|
Root directory of the LeRobot dataset. |