Source code for schola.core.spaces.box
Copied!
# Copyright (c) 2024 Advanced Micro Devices, Inc. All Rights Reserved.
"""
Implementation of a BoxSpace, a space representing a bounded vector of continuous values.
"""
from typing import Dict, List, Optional, Tuple, Union
import gymnasium
import schola.generated.Spaces_pb2 as proto_spaces
import schola.generated.Points_pb2 as proto_points
import numpy as np
from .base import UnrealSpace
[docs]
class BoxSpace(gymnasium.spaces.Box, UnrealSpace):
"""
A Space representing a box in n-dimensional space.
Parameters
----------
low : Union[float, np.ndarray, List[float]]
The lower bounds of the box.
high : Union[float, np.ndarray, List[float]]
The upper bounds of the box.
shape : Tuple[int], optional
The shape of the space.
Attributes
----------
shape : Tuple[int]
The shape of the space.
Note
----
Unlike, the gymnasium Box space, this class does not have a dtype attribute. The dtype is always np.float32.
See Also
--------
gymnasium.spaces.Box : The gym space object that this class is analogous to.
proto_spaces.BoxSpace : The protobuf representation of this space.
"""
proto_space = proto_spaces.BoxSpace
_name = "box_space"
[docs]
def __init__(self, low:Union[float, np.ndarray,List[float]], high:Union[float, np.ndarray,List[float]],shape:Optional[Tuple[int]]=None):
if isinstance(low, list):
low = np.asarray(low, dtype=np.float32)
if isinstance(high, list):
high = np.asarray(high, dtype=np.float32)
super().__init__(low=low, high=high,shape=shape)
[docs]
@classmethod
def from_proto(cls, message : proto_spaces.BoxSpace) -> "BoxSpace":
low = []
high = []
for dimension in message.dimensions:
low.append(dimension.low)
high.append(dimension.high)
return BoxSpace(low=low, high=high)
[docs]
@classmethod
def is_empty_definition(cls, message : proto_spaces.BoxSpace) -> bool:
return len(list(message.dimensions)) == 0
[docs]
def fill_proto(self, msg: proto_points.FundamentalPoint, values):
msg.box_point.values.extend(values)
[docs]
def to_normalized(self):
"""
Normalize the bounds of the space to be between 0 and 1
Returns
-------
BoxSpace
The normalized space. A modified version of the space this method is called on
Examples
--------
>>> space = BoxSpace([0, 0],[2, 2])
>>> space.to_normalized() == BoxSpace([0., 0.], [1., 1.])
True
"""
self.low = np.asarray([0. for x in self.low], dtype=np.float32)
self.high = np.asarray([1. for x in self.high], dtype=np.float32)
return self
[docs]
@classmethod
def merge(cls,*spaces: List["BoxSpace"]) -> "BoxSpace":
"""
Merge multiple BoxSpaces into a single space.
Parameters
----------
*spaces : List[BoxSpace]
The spaces to merge.
Returns
-------
BoxSpace
The merged space.
Raises
------
TypeError
If any of the spaces are not BoxSpaces.
Examples
--------
>>> merged_space = BoxSpace.merge(BoxSpace([0,0],[1,1]), BoxSpace([2,2],[3,3]))
>>> merged_space == BoxSpace([0, 0, 2, 2], [1, 1, 3, 3])
True
"""
for space in spaces:
if not isinstance(space,gymnasium.spaces.Box):
raise TypeError(f"Cannot merge BoxSpace with {type(space)}")
low = np.concatenate([space.low for space in spaces])
high = np.concatenate([space.high for space in spaces])
return BoxSpace(low,high)
def __len__(self) -> int:
"""
Get the number of dimensions of the space
Returns
-------
int
The number of dimensions of the space
Examples
--------
>>> space = BoxSpace([0,0],[1,1])
>>> len(space)
2
"""
return self.shape[0]
[docs]
def process_data(self, msg : proto_points.FundamentalPoint) -> np.ndarray:
return np.asarray(msg.box_point.values)
