from __future__ import annotations
from typing import Any, Iterable, Iterator, Mapping, cast
import json
from collections import OrderedDict
from pathlib import Path
import numpy as np
from ConfigSpace import Configuration, ConfigurationSpace
from smac.constants import MAXINT
from smac.multi_objective.abstract_multi_objective_algorithm import (
AbstractMultiObjectiveAlgorithm,
)
from smac.runhistory.dataclasses import (
InstanceSeedBudgetKey,
InstanceSeedKey,
TrialInfo,
TrialKey,
TrialValue,
)
from smac.runhistory.enumerations import StatusType
from smac.utils.configspace import get_config_hash
from smac.utils.logging import get_logger
from smac.utils.multi_objective import normalize_costs
__copyright__ = "Copyright 2022, automl.org"
__license__ = "3-clause BSD"
logger = get_logger(__name__)
[docs]class RunHistory(Mapping[TrialKey, TrialValue]):
"""Container for the target function run information.
Most importantly, the runhistory contains an efficient mapping from each evaluated configuration to the
empirical cost observed on either the full instance set or a subset. The cost is the average over all
observed costs for one configuration:
* If using budgets for a single instance, only the cost on the highest observed budget is returned.
* If using instances as the budget, the average cost over all evaluated instances is returned.
* Theoretically, the runhistory object can handle instances and budgets at the same time. This is
neither used nor tested.
Note
----
Guaranteed to be picklable.
Parameters
----------
multi_objective_algorithm : AbstractMultiObjectiveAlgorithm | None, defaults to None
The multi-objective algorithm is required to scalarize the costs in case of multi-objective.
overwrite_existing_trials : bool, defaults to false
Overwrites a trial (combination of configuration, instance, budget and seed) if it already exists.
"""
def __init__(
self,
multi_objective_algorithm: AbstractMultiObjectiveAlgorithm | None = None,
overwrite_existing_trials: bool = False,
) -> None:
self._multi_objective_algorithm = multi_objective_algorithm
self._overwrite_existing_trials = overwrite_existing_trials
self.reset()
@property
def submitted(self) -> int:
"""Returns how many trials have been submitted."""
return self._submitted
@property
def finished(self) -> int:
"""Returns how many trials have been finished."""
return self._finished
@property
def running(self) -> int:
"""Returns how many trials are still running."""
return self._running
@property
def multi_objective_algorithm(self) -> AbstractMultiObjectiveAlgorithm | None:
"""The multi-objective algorithm required to scaralize the costs in case of multi-objective."""
return self._multi_objective_algorithm
@multi_objective_algorithm.setter
def multi_objective_algorithm(self, value: AbstractMultiObjectiveAlgorithm) -> None:
"""We want to have the option to change the multi objective algorithm."""
self._multi_objective_algorithm = value
@property
def ids_config(self) -> dict[int, Configuration]:
"""Mapping from config id to configuration."""
return self._ids_config
@property
def config_ids(self) -> dict[Configuration, int]:
"""Mapping from configuration to config id."""
return self._config_ids
@property
def objective_bounds(self) -> list[tuple[float, float]]:
"""Returns the lower and upper bound of each objective."""
return self._objective_bounds
[docs] def reset(self) -> None:
"""Resets this runhistory to its default state."""
# By having the data in a deterministic order we can do useful tests when we
# serialize the data and can assume it is still in the same order as it was added.
self._data: dict[TrialKey, TrialValue] = OrderedDict()
# Keep track of trials
self._submitted = 0
self._finished = 0
self._running = 0
# For fast access, we have also an unordered data structure to get all instance
# seed pairs of a configuration.
self._config_id_to_isk_to_budget: dict[int, dict[InstanceSeedKey, list[float | None]]] = {}
self._running_trials: list[TrialInfo] = []
self._config_ids: dict[Configuration, int] = {}
self._ids_config: dict[int, Configuration] = {}
self._n_id = 0
# Stores cost for each configuration ID
self._cost_per_config: dict[int, float | list[float]] = {}
# Stores min cost across all budgets for each configuration ID
self._min_cost_per_config: dict[int, float | list[float]] = {}
# Maps the configuration ID to the number of runs for that configuration
# and is necessary for computing the moving average.
self._num_trials_per_config: dict[int, int] = {}
# Store whether a datapoint is "external", which means it was read from
# a JSON file. Can be chosen to not be written to disk.
self._n_objectives: int = -1
self._objective_bounds: list[tuple[float, float]] = []
[docs] def __contains__(self, k: object) -> bool:
"""Dictionary semantics for `k in runhistory`."""
return k in self._data
[docs] def __getitem__(self, k: TrialKey) -> TrialValue:
"""Dictionary semantics for `v = runhistory[k]`."""
return self._data[k]
[docs] def __iter__(self) -> Iterator[TrialKey]:
"""Dictionary semantics for `for k in runhistory.keys()`."""
return iter(self._data.keys())
[docs] def __len__(self) -> int:
"""Enables the `len(runhistory)`"""
return len(self._data)
[docs] def __eq__(self, other: Any) -> bool:
"""Enables to check equality of runhistory if the run is continued."""
return self._data == other._data
[docs] def empty(self) -> bool:
"""Check whether the RunHistory is empty.
Returns
-------
emptiness: bool
True if trials have been added to the RunHistory.
"""
return len(self._data) == 0
[docs] def add(
self,
config: Configuration,
cost: int | float | list[int | float],
time: float = 0.0,
status: StatusType = StatusType.SUCCESS,
instance: str | None = None,
seed: int | None = None,
budget: float | None = None,
starttime: float = 0.0,
endtime: float = 0.0,
additional_info: dict[str, Any] = {},
force_update: bool = False,
) -> None:
"""Adds a new trial to the RunHistory.
Parameters
----------
config : Configuration
cost : int | float | list[int | float]
Cost of the evaluated trial. Might be a list in case of multi-objective.
time : float
How much time was needed to evaluate the trial.
status : StatusType, defaults to StatusType.SUCCESS
The status of the trial.
instance : str | None, defaults to none
seed : int | None, defaults to none
budget : float | None, defaults to none
starttime : float, defaults to 0.0
endtime : float, defaults to 0.0
additional_info : dict[str, Any], defaults to {}
force_update : bool, defaults to false
Overwrites a previous trial if the trial already exists.
"""
if config is None:
raise TypeError("Configuration must not be None.")
elif not isinstance(config, Configuration):
raise TypeError("Configuration is not of type Configuration, but %s." % type(config))
# Squeeze is important to reduce arrays with one element
# to scalars.
cost_array = np.asarray(cost).squeeze()
n_objectives = np.size(cost_array)
# Get the config id
config_id = self._config_ids.get(config)
if config_id is None:
self._n_id += 1
self._config_ids[config] = self._n_id
self._ids_config[self._n_id] = config
config_id = self._n_id
# Set the id attribute of the config object, so that users can access it
config.config_id = config_id
if status != StatusType.RUNNING:
if self._n_objectives == -1:
self._n_objectives = n_objectives
elif self._n_objectives != n_objectives:
raise ValueError(
f"Cost is not of the same length ({n_objectives}) as the number of "
f"objectives ({self._n_objectives})."
)
# Let's always work with floats; Makes it easier to deal with later on
# array.tolist(), it returns a scalar if the array has one element.
c = cost_array.tolist()
if self._n_objectives == 1:
c = float(c)
else:
c = [float(i) for i in c]
else:
c = cost_array.tolist()
if budget is not None:
# Just to make sure we really add a float
budget = float(budget)
k = TrialKey(config_id=config_id, instance=instance, seed=seed, budget=budget)
v = TrialValue(
cost=c,
time=time,
status=status,
starttime=starttime,
endtime=endtime,
additional_info=additional_info,
)
# Construct keys and values for the data dictionary
for key, value in (
("config", config.get_dictionary()),
("config_id", config_id),
("instance", instance),
("seed", seed),
("budget", budget),
("cost", c),
("time", time),
("status", status),
("starttime", starttime),
("endtime", endtime),
("additional_info", additional_info),
("origin", config.origin),
):
self._check_json_serializable(key, value, k, v)
# Each trial_key is supposed to be used only once. Repeated tries to add
# the same trial_key will be ignored silently if not capped.
previous_k = self._data.get(k)
if self._overwrite_existing_trials or force_update or previous_k is None:
# Update stati
if previous_k is None:
if status == StatusType.RUNNING:
self._running += 1
else:
self._finished += 1
self._submitted += 1
else:
if previous_k.status == StatusType.RUNNING and status != StatusType.RUNNING:
self._running -= 1
self._finished += 1
self._add(k, v, status)
else:
logger.info("Entry was not added to the runhistory because existing trials will not be overwritten.")
[docs] def add_trial(self, info: TrialInfo, value: TrialValue) -> None:
"""Adds a trial to the runhistory.
Parameters
----------
trial : TrialInfo
The ``TrialInfo`` object of the running trial.
"""
self.add(
config=info.config,
cost=value.cost,
time=value.time,
status=value.status,
instance=info.instance,
seed=info.seed,
budget=info.budget,
starttime=value.starttime,
endtime=value.endtime,
additional_info=value.additional_info,
)
[docs] def add_running_trial(self, trial: TrialInfo) -> None:
"""Adds a running trial to the runhistory.
Parameters
----------
trial : TrialInfo
The ``TrialInfo`` object of the running trial.
"""
self.add(
config=trial.config,
cost=float(MAXINT),
time=0.0,
status=StatusType.RUNNING,
instance=trial.instance,
seed=trial.seed,
budget=trial.budget,
)
[docs] def update_cost(self, config: Configuration) -> None:
"""Stores the performance of a configuration across the instances in `self._cost_per_config`
and also updates `self._num_trials_per_config`.
Parameters
----------
config: Configuration
configuration to update cost based on all trials in runhistory
"""
config_id = self._config_ids[config]
# Removing duplicates while keeping the order
inst_seed_budgets = list(
dict.fromkeys(self.get_instance_seed_budget_keys(config, highest_observed_budget_only=True))
)
self._cost_per_config[config_id] = self.average_cost(config, inst_seed_budgets)
self._num_trials_per_config[config_id] = len(inst_seed_budgets)
all_isb = list(dict.fromkeys(self.get_instance_seed_budget_keys(config, highest_observed_budget_only=False)))
self._min_cost_per_config[config_id] = self.min_cost(config, all_isb)
[docs] def incremental_update_cost(self, config: Configuration, cost: float | list[float]) -> None:
"""Incrementally updates the performance of a configuration by using a moving average.
Parameters
----------
config: Configuration
configuration to update cost based on all trials in runhistory
cost: float
cost of new run of config
"""
config_id = self._config_ids[config]
n_trials = self._num_trials_per_config.get(config_id, 0)
if self._n_objectives > 1:
costs = np.array(cost)
old_costs = self._cost_per_config.get(config_id, np.array([0.0 for _ in range(self._n_objectives)]))
old_costs = np.array(old_costs)
new_costs = ((old_costs * n_trials) + costs) / (n_trials + 1)
self._cost_per_config[config_id] = new_costs.tolist()
else:
old_cost = self._cost_per_config.get(config_id, 0.0)
assert isinstance(cost, float)
assert isinstance(old_cost, float)
self._cost_per_config[config_id] = ((old_cost * n_trials) + cost) / (n_trials + 1)
self._num_trials_per_config[config_id] = n_trials + 1
[docs] def get_cost(self, config: Configuration) -> float:
"""Returns empirical cost for a configuration. See the class docstring for how the costs are
computed. The costs are not re-computed, but are read from cache.
Parameters
----------
config: Configuration
Returns
-------
cost: float
Computed cost for configuration
"""
config_id = self._config_ids.get(config)
# Cost is always a single value (Single objective) or a list of values (Multi-objective)
# For example, _cost_per_config always holds the value on the highest budget
cost = self._cost_per_config.get(config_id, np.nan) # type: ignore[arg-type] # noqa F821
if self._n_objectives > 1:
assert isinstance(cost, list)
assert self.multi_objective_algorithm is not None
# We have to normalize the costs here
costs = normalize_costs(cost, self._objective_bounds)
# After normalization, we get the weighted average
return self.multi_objective_algorithm(costs)
assert isinstance(cost, float)
return float(cost)
[docs] def get_min_cost(self, config: Configuration) -> float:
"""Returns the lowest empirical cost for a configuration across all trials.
See the class docstring for how the costs are computed. The costs are not re-computed
but are read from cache.
Parameters
----------
config : Configuration
Returns
-------
min_cost: float
Computed cost for configuration
"""
config_id = self._config_ids.get(config)
cost = self._min_cost_per_config.get(config_id, np.nan) # type: ignore
if self._n_objectives > 1:
assert type(cost) == list
assert self.multi_objective_algorithm is not None
costs = normalize_costs(cost, self._objective_bounds)
# Note: We have to mean here because we already got the min cost
return self.multi_objective_algorithm(costs)
assert type(cost) == float
return float(cost)
[docs] def average_cost(
self,
config: Configuration,
instance_seed_budget_keys: list[InstanceSeedBudgetKey] | None = None,
normalize: bool = False,
) -> float | list[float]:
"""Return the average cost of a configuration. This is the mean of costs of all instance-
seed pairs.
Parameters
----------
config : Configuration
Configuration to calculate objective for.
instance_seed_budget_keys : list, optional (default=None)
List of tuples of instance-seeds-budget keys. If None, the runhistory is
queried for all trials of the given configuration.
normalize : bool, optional (default=False)
Normalizes the costs wrt. objective bounds in the multi-objective setting.
Only a float is returned if normalize is True. Warning: The value can change
over time because the objective bounds are changing. Also, the objective weights are
incorporated.
Returns
-------
Cost: float | list[float]
Average cost. In case of multiple objectives, the mean of each objective is returned.
"""
costs = self._cost(config, instance_seed_budget_keys)
if costs:
if self._n_objectives > 1:
# Each objective is averaged separately
# [[100, 200], [0, 0]] -> [50, 100]
averaged_costs = np.mean(costs, axis=0).tolist()
if normalize:
assert self.multi_objective_algorithm is not None
normalized_costs = normalize_costs(averaged_costs, self._objective_bounds)
return self.multi_objective_algorithm(normalized_costs)
else:
return averaged_costs
return float(np.mean(costs))
return np.nan
[docs] def sum_cost(
self,
config: Configuration,
instance_seed_budget_keys: list[InstanceSeedBudgetKey] | None = None,
normalize: bool = False,
) -> float | list[float]:
"""Return the sum of costs of a configuration. This is the sum of costs of all instance-seed
pairs.
Parameters
----------
config : Configuration
Configuration to calculate objective for.
instance_seed_budget_keys : list, optional (default=None)
List of tuples of instance-seeds-budget keys. If None, the runhistory is
queried for all trials of the given configuration.
normalize : bool, optional (default=False)
Normalizes the costs wrt objective bounds in the multi-objective setting.
Only a float is returned if normalize is True. Warning: The value can change
over time because the objective bounds are changing. Also, the objective weights are
incorporated.
Returns
-------
sum_cost: float | list[float]
Sum of costs of config. In case of multiple objectives, the costs are summed up for each
objective individually.
"""
costs = self._cost(config, instance_seed_budget_keys)
if costs:
if self._n_objectives > 1:
# Each objective is summed separately
# [[100, 200], [20, 10]] -> [120, 210]
summed_costs = np.sum(costs, axis=0).tolist()
if normalize:
assert self.multi_objective_algorithm is not None
normalized_costs = normalize_costs(summed_costs, self._objective_bounds)
return self.multi_objective_algorithm(normalized_costs)
else:
return summed_costs
return float(np.sum(costs))
[docs] def min_cost(
self,
config: Configuration,
instance_seed_budget_keys: list[InstanceSeedBudgetKey] | None = None,
normalize: bool = False,
) -> float | list[float]:
"""Return the minimum cost of a configuration. This is the minimum cost of all instance-seed
pairs.
Warning
-------
In the case of multi-fidelity, the minimum cost per objectives is returned.
Parameters
----------
config : Configuration
Configuration to calculate objective for.
instance_seed_budget_keys : list, optional (default=None)
List of tuples of instance-seeds-budget keys. If None, the runhistory is
queried for all trials of the given configuration.
normalize : bool, optional (default=False)
Normalizes the costs wrt objective bounds in the multi-objective setting.
Only a float is returned if normalize is True. Warning: The value can change
over time because the objective bounds are changing. Also, the objective weights are
incorporated.
Returns
-------
min_cost: float | list[float]
Minimum cost of the config. In case of multi-objective, the minimum cost per objective
is returned.
"""
costs = self._cost(config, instance_seed_budget_keys)
if costs:
if self._n_objectives > 1:
# Each objective is viewed separately
# [[100, 200], [20, 500]] -> [20, 200]
min_costs = np.min(costs, axis=0).tolist()
if normalize:
assert self.multi_objective_algorithm is not None
normalized_costs = normalize_costs(min_costs, self._objective_bounds)
return self.multi_objective_algorithm(normalized_costs)
else:
return min_costs
return float(np.min(costs))
return np.nan
[docs] def get_config(self, config_id: int) -> Configuration:
"""Returns the configuration from the configuration id."""
return self._ids_config[config_id]
[docs] def get_config_id(self, config: Configuration) -> int:
"""Returns the configuration id from a configuration."""
return self._config_ids[config]
[docs] def get_configs(self, sort_by: str | None = None) -> list[Configuration]:
"""Return all configurations in this RunHistory object.
Parameters
----------
sort_by : str | None, defaults to None
Sort the configs by ``cost`` (lowest cost first) or ``num_trials`` (config with lowest number of trials
first).
Returns
-------
configurations : list
All configurations in the runhistory.
"""
configs = list(self._config_ids.keys())
if sort_by == "cost":
return sorted(configs, key=lambda config: self._cost_per_config[self._config_ids[config]])
elif sort_by == "num_trials":
return sorted(configs, key=lambda config: len(self.get_trials(config)))
elif sort_by is None:
return configs
else:
raise ValueError(f"Unknown sort_by value: {sort_by}.")
[docs] def get_configs_per_budget(
self,
budget_subset: list[float | int | None] | None = None,
) -> list[Configuration]:
"""Return all configs in this runhistory that have been run on one of these budgets.
Parameters
----------
budget_subset: list[float | int | None] | None, defaults to None
Returns
-------
configurations : list
List of configurations that have been run on the budgets in ``budget_subset``.
"""
if budget_subset is None:
return self.get_configs()
configs = []
for key in self._data.keys():
if key.budget in budget_subset:
configs.append(self._ids_config[key.config_id])
return configs
[docs] def get_running_configs(self) -> list[Configuration]:
"""Returns all configurations which have at least one running trial.
Returns
-------
list[Configuration]
List of configurations, all of which have at least one running trial.
"""
configs = []
for trial in self._running_trials:
if trial.config not in configs:
configs.append(trial.config)
return configs
[docs] def get_trials(
self,
config: Configuration,
highest_observed_budget_only: bool = True,
) -> list[TrialInfo]:
"""Returns all trials for a configuration.
Warning
-------
Does not return running trials. Please use ``get_running_trials`` to receive running trials.
Parameters
----------
config : Configuration
highest_observed_budget_only : bool
Select only the highest observed budget run for this configuration.
Meaning on multiple executions of the same instance-seed pair for a
a given configuration, only the highest observed budget is returned.
Returns
-------
trials : list[InstanceSeedBudgetKey]
List of trials for the passed configuration.
"""
config_id = self._config_ids.get(config)
trials = {}
if config_id in self._config_id_to_isk_to_budget:
trials = self._config_id_to_isk_to_budget[config_id].copy()
# Select only the max budget run if specified
if highest_observed_budget_only:
for k, v in trials.items():
if None in v:
trials[k] = [None]
else:
trials[k] = [max([v_ for v_ in v if v_ is not None])]
return [TrialInfo(config, k.instance, k.seed, budget) for k, v in trials.items() for budget in v]
[docs] def get_running_trials(self, config: Configuration | None = None) -> list[TrialInfo]:
"""Returns all running trials for the passed configuration.
Parameters
----------
config : Configuration | None, defaults to None
Return only running trials from the passed configuration. If None, all configs are
considered.
Returns
-------
trials : list[TrialInfo]
List of trials, all of which are still running.
"""
# Always work on copies
if config is None:
return [trial for trial in self._running_trials]
else:
return [trial for trial in self._running_trials if trial.config == config]
[docs] def get_instance_seed_budget_keys(
self,
config: Configuration,
highest_observed_budget_only: bool = True,
) -> list[InstanceSeedBudgetKey]:
"""
Uses ``get_trials`` to return a list of instance-seed-budget keys.
Warning
-------
Does not return running instances.
Parameters
----------
config : Configuration
highest_observed_budget_only : bool, defaults to True
Select only the highest observed budget run for this configuration.
Returns
-------
list[InstanceSeedBudgetKey]
"""
trials = self.get_trials(config, highest_observed_budget_only)
# Convert to instance-seed-budget key
return [InstanceSeedBudgetKey(t.instance, t.seed, t.budget) for t in trials]
[docs] def save(self, filename: str | Path = "runhistory.json") -> None:
"""Saves RunHistory to disk.
Parameters
----------
filename : str | Path, defaults to "runhistory.json"
"""
data = []
for k, v in self._data.items():
data += [
(
int(k.config_id),
str(k.instance) if k.instance is not None else None,
int(k.seed) if k.seed is not None else None,
float(k.budget) if k.budget is not None else None,
v.cost,
v.time,
v.status,
v.starttime,
v.endtime,
v.additional_info,
)
]
config_ids_to_serialize = set([entry[0] for entry in data])
configs = {}
config_origins = {}
for id_, config in self._ids_config.items():
if id_ in config_ids_to_serialize:
configs[id_] = config.get_dictionary()
config_origins[id_] = config.origin
if isinstance(filename, str):
filename = Path(filename)
assert str(filename).endswith(".json")
filename.parent.mkdir(parents=True, exist_ok=True)
with open(filename, "w") as fp:
assert self._running == len(self._running_trials)
json.dump(
{
"stats": {"submitted": self._submitted, "finished": self._finished, "running": self._running},
"data": data,
"configs": configs,
"config_origins": config_origins,
},
fp,
indent=2,
)
[docs] def load(self, filename: str | Path, configspace: ConfigurationSpace) -> None:
"""Loads the runhistory from disk.
Warning
-------
Overwrites the current runhistory.
Parameters
----------
filename : str | Path
configspace : ConfigSpace
"""
if isinstance(filename, str):
filename = Path(filename)
# We reset the RunHistory first to avoid any inconsistencies
self.reset()
try:
with open(filename) as fp:
data = json.load(fp)
except Exception as e:
logger.warning(
f"Encountered exception {e} while reading RunHistory from {filename}. Not adding any trials!"
)
return
config_origins = data.get("config_origins", {})
self._ids_config = {}
for id_, values in data["configs"].items():
self._ids_config[int(id_)] = Configuration(
configspace,
values=values,
origin=config_origins.get(id_, None),
)
self._config_ids = {config: id_ for id_, config in self._ids_config.items()}
self._n_id = len(self._config_ids)
# Important to use add method to use all data structure correctly
for entry in data["data"]:
# Set n_objectives first
if self._n_objectives == -1:
if isinstance(entry[4], float) or isinstance(entry[4], int):
self._n_objectives = 1
else:
self._n_objectives = len(entry[4])
cost: list[float] | float
if self._n_objectives == 1:
cost = float(entry[4])
else:
cost = [float(x) for x in entry[4]]
self.add(
config=self._ids_config[int(entry[0])],
cost=cost,
time=float(entry[5]),
status=StatusType(entry[6]),
instance=entry[1],
seed=entry[2],
budget=entry[3],
starttime=entry[7],
endtime=entry[8],
additional_info=entry[9],
)
# Although adding trials should give us the same stats, the trajectory might be different
# because of the running status and/or overwriting trials
# Therefore, we just overwrite them
self._submitted = data["stats"]["submitted"]
self._finished = data["stats"]["finished"]
self._running = data["stats"]["running"]
[docs] def update_from_json(
self,
filename: str,
configspace: ConfigurationSpace,
) -> None:
"""Updates the current RunHistory by adding new trials from a json file.
Parameters
----------
filename : str
File name to load from.
configspace : ConfigurationSpace
"""
new_runhistory = RunHistory()
new_runhistory.load(filename, configspace)
self.update(runhistory=new_runhistory)
[docs] def update(self, runhistory: RunHistory) -> None:
"""Updates the current RunHistory by adding new trials from another RunHistory.
Parameters
----------
runhistory : RunHistory
RunHistory with additional data to be added to self
"""
# Configurations might be already known, but by a different ID. This
# does not matter here because the add() method handles this
# correctly by assigning an ID to unknown configurations and re-using the ID.
for key, value in runhistory.items():
config = runhistory._ids_config[key.config_id]
self.add(
config=config,
cost=value.cost,
time=value.time,
status=value.status,
instance=key.instance,
starttime=value.starttime,
endtime=value.endtime,
seed=key.seed,
budget=key.budget,
additional_info=value.additional_info,
)
[docs] def update_costs(self, instances: list[str] | None = None) -> None:
"""Computes the cost of all configurations from scratch and overwrites `self._cost_per_config`
and `self._num_trials_per_config` accordingly.
Parameters
----------
instances: list[str] | None, defaults to none
List of instances; if given, cost is only computed wrt to this instance set.
"""
self._cost_per_config = {}
self._num_trials_per_config = {}
for config, config_id in self._config_ids.items():
# Removing duplicates while keeping the order
inst_seed_budgets = list(
dict.fromkeys(self.get_instance_seed_budget_keys(config, highest_observed_budget_only=True))
)
if instances is not None:
inst_seed_budgets = list(filter(lambda x: x.instance in cast(list, instances), inst_seed_budgets))
if inst_seed_budgets: # can be empty if never saw any trials on instances
self._cost_per_config[config_id] = self.average_cost(config, inst_seed_budgets)
self._min_cost_per_config[config_id] = self.min_cost(config, inst_seed_budgets)
self._num_trials_per_config[config_id] = len(inst_seed_budgets)
def _check_json_serializable(
self,
key: str,
obj: Any,
trial_key: TrialKey,
trial_value: TrialValue,
) -> None:
try:
json.dumps(obj)
except Exception as e:
raise ValueError(
"Cannot add %s: %s of type %s to runhistory because it raises an error during JSON encoding, "
"please see the error above.\ntrial_key: %s\ntrial_value %s"
% (key, str(obj), type(obj), trial_key, trial_value)
) from e
def _update_objective_bounds(self) -> None:
"""Update the objective bounds based on the data in the RunHistory."""
all_costs = []
for run_value in self._data.values():
costs = run_value.cost
if run_value.status == StatusType.SUCCESS:
if not isinstance(costs, Iterable):
costs = [costs]
assert len(costs) == self._n_objectives
all_costs.append(costs)
all_costs = np.array(all_costs, dtype=float) # type: ignore[assignment]
if len(all_costs) == 0:
self._objective_bounds = [(np.inf, -np.inf)] * self._n_objectives
return
min_values = np.min(all_costs, axis=0)
max_values = np.max(all_costs, axis=0)
self._objective_bounds = []
for min_v, max_v in zip(min_values, max_values):
self._objective_bounds += [(min_v, max_v)]
def _add(self, k: TrialKey, v: TrialValue, status: StatusType) -> None:
"""
Actual function to add new entry to data structures.
Note
----
This method always calls `update_cost` in the multi-objective setting.
"""
self._data[k] = v
# Update objective bounds based on raw data
self._update_objective_bounds()
# Do not register the cost until the run has completed
if status != StatusType.RUNNING:
# Also add to fast data structure
isk = InstanceSeedKey(k.instance, k.seed)
self._config_id_to_isk_to_budget[k.config_id] = self._config_id_to_isk_to_budget.get(k.config_id, {})
# We sanity-check whether we don't mix none and str in the instances
for isk_ in self._config_id_to_isk_to_budget[k.config_id].keys():
if isinstance(isk_, str) != isinstance(isk, str):
raise ValueError(
"Can not mix instances of different types. "
f"Wants to add {isk_.instance} but found already {isk.instance}."
)
if isk not in self._config_id_to_isk_to_budget[k.config_id]:
# Add new inst-seed-key with budget to main dict
self._config_id_to_isk_to_budget[k.config_id][isk] = [k.budget]
# Before it was k.budget not in isk
elif k.budget != isk.instance and k.budget != isk.seed:
# We have to make sure that we don't mix none and float budgets
if isinstance(self._config_id_to_isk_to_budget[k.config_id][isk][0], float) != isinstance(
k.budget, float
):
raise ValueError(
"Can not mix budgets of different types for the same instance-seed pair. "
f"Wants to add {k.budget} but found already "
f"{self._config_id_to_isk_to_budget[k.config_id][isk][0]}."
)
# Append new budget to existing inst-seed-key dict
self._config_id_to_isk_to_budget[k.config_id][isk].append(k.budget)
config = self._ids_config[k.config_id]
config_hash = get_config_hash(config)
# If budget is used, then update cost instead of incremental updates
if not self._overwrite_existing_trials and k.budget == 0:
logger.debug(f"Incremental update cost for config {config_hash}")
# Assumes an average across trials as cost function aggregation, this is used for
# algorithm configuration (incremental updates are used to save time as getting the
# cost for > 100 instances is high)
self.incremental_update_cost(config, v.cost)
else:
# This happens when budget > 0 (only successive halving and hyperband so far)
logger.debug(f"Update cost for config {config_hash}.")
self.update_cost(config)
# Make TrialInfo object
trial_info = TrialInfo(self.get_config(k.config_id), instance=k.instance, seed=k.seed, budget=k.budget)
# Fast data structure for pending trials
if status == StatusType.RUNNING:
# Add to running cache
self._running_trials.append(trial_info)
else:
# Remove from cache
if trial_info in self._running_trials:
self._running_trials.remove(trial_info)
def _cost(
self,
config: Configuration,
instance_seed_budget_keys: list[InstanceSeedBudgetKey] | None = None,
) -> list[float | list[float]]:
"""Returns a list of all costs for the given config for further calculations.
The costs are directly taken from the RunHistory data.
Parameters
----------
config : Configuration
Configuration to calculate objective for.
instance_seed_budget_keys : list, defaults to None
List of tuples of instance-seeds-budget keys. If None, the RunHistory is
queried for all trials of the given configuration.
Returns
-------
costs: list[list[float] | list[list[float]]]
List of all found costs. In case of multi-objective, the list contains lists.
"""
try:
id_ = self._config_ids[config]
except KeyError: # Challenger was not running so far
return []
if instance_seed_budget_keys is None:
instance_seed_budget_keys = self.get_instance_seed_budget_keys(config, highest_observed_budget_only=True)
costs = []
for key in instance_seed_budget_keys:
k = TrialKey(
config_id=id_,
instance=key.instance,
seed=key.seed,
budget=key.budget,
)
costs.append(self._data[k].cost)
return costs