#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Author : qichun tang
# @Date : 2020-12-18
# @Contact : qichun.tang@bupt.edu.cn
from typing import Union
import numpy as np
from joblib import load
from ultraopt.optimizer.base_opt import BaseOptimizer
from ultraopt.utils.config_space import get_dict_from_config
[文档]def warm_start_optimizer(optimizer: BaseOptimizer, previous_result,
warm_start_strategy="resume"):
from ultraopt.facade.result import FMinResult
if previous_result is None:
return optimizer
if isinstance(previous_result, str):
previous_result = load(previous_result) # type: Union[FMinResult, BaseOptimizer]
if warm_start_strategy == "resume":
for budget, obvs in previous_result.budget2obvs.items():
L = len(obvs["losses"])
for i in range(L):
optimizer.tell(
get_dict_from_config(obvs["configs"][i]),
obvs["losses"][i],
budget,
update_model=(i == L - 1))
elif warm_start_strategy == "continue":
if isinstance(previous_result, FMinResult):
prev_opt = previous_result.optimizer
else:
prev_opt = previous_result
prev_opt.resume_time()
return prev_opt
else:
raise ValueError(f"Invalid warm_start_strategy {warm_start_strategy} not in ['resume', 'continue']")
return optimizer
[文档]def get_wanted(opt_: BaseOptimizer):
budget2obvs = opt_.budget2obvs
max_budget = opt_.get_available_max_budget()
obvs = budget2obvs[max_budget]
losses = obvs["losses"]
configs = obvs["configs"]
if len(losses):
idx = np.argmin(losses)
best_loss = losses[idx]
best_config = get_dict_from_config(configs[idx])
return max_budget, best_loss, best_config
return max_budget, np.inf, None
