asp技校网站优化教程网
# 获取训练目标值,计算Pareto前沿(非支配解集合),然后从样本中提取出Pareto最优解。train_obj = self.samples[1]pareto_mask = is_non_dominated(train_obj)pareto_y = train_obj[pareto_mask]
源码
这里用到了一个函数 is_non_dominated(),来看下该函数的源码:
from __future__ import annotationsimport torch
from torch import Tensordef is_non_dominated(Y: Tensor, deduplicate: bool = True) -> Tensor:r"""Computes the non-dominated front.Note: this assumes maximization.Args:输入:张量 Y,其维度为 (batch_shape) x n x m,这里 n 代表样本数量,m 代表每个样本的目标数量。,其中括号括住的batch_shape意思是可选,可以有这个维度或者没有Y: A `(batch_shape) x n x m`-dim tensor of outcomes.deduplicate: A boolean indicating whether to only return unique points on the pareto frontier.Returns:返回:布尔张量,指示每个样本是否是非支配点A `(batch_shape) x n`-dim boolean tensor indicating whether each point is non-dominated."""Y1 = Y.unsqueeze(-3)Y2 = Y.unsqueeze(-2)dominates = (Y1 >= Y2).all(dim=-1) & (Y1 > Y2).any(dim=-1)nd_mask = ~(dominates.any(dim=-1))if deduplicate:# remove duplicates# find index of first occurrence of each unique elementindices = (Y1 == Y2).all(dim=-1).long().argmax(dim=-1)keep = torch.zeros_like(nd_mask)keep.scatter_(dim=-1, index=indices, value=1.0)return nd_mask & keepreturn nd_mask
示例:
有一组解,每个解有两个目标值,找出这组解中的非支配解:
import torch
from botorch.utils.multi_objective import is_non_dominated# 假设有5个解,每个解有2个目标
Y = torch.tensor([[0.5, 0.7],[0.6, 0.6],[0.8, 0.3],[0.4, 0.9],[0.7, 0.5]
])# 调用 is_non_dominated 函数
non_dominated_mask = is_non_dominated(Y)# 打印非支配解
print("Non-dominated solutions:", Y[non_dominated_mask])
实例:
这里就是先拿到所有的目标值,然后计算哪些是Pareto(True或者False),最后再原始数据中选出所有True的数据
# 获取训练目标值,计算Pareto前沿(非支配解集合),然后从样本中提取出Pareto最优解。train_obj = self.samples[1]pareto_mask = is_non_dominated(train_obj)pareto_y = train_obj[pareto_mask]