Continue an Optimization

SMAC can also be continued from a previous run. To do so, it reads in old files (derived from scenario’s name, output_directory and seed) and sets the corresponding components. In this example, an optimization of a simple quadratic function is continued.

First, after creating a scenario with 50 trials, we run SMAC with overwrite=True. This will overwrite any previous runs (in case the example was called before). We use a custom callback to artificially stop this first optimization after 10 trials.

Second, we again run the SMAC optimization using the same scenario, but this time with overwrite=False. As there already is a previous run with the same meta data, this run will be continued until the 50 trials are reached.

[INFO][abstract_initial_design.py:147] Using 10 initial design configurations and 0 additional configurations.
[INFO][abstract_intensifier.py:306] Using only one seed for deterministic scenario.
[INFO][abstract_intensifier.py:516] Added config f09c3b as new incumbent because there are no incumbents yet.
[INFO][abstract_intensifier.py:595] Added config bec0fc and rejected config f09c3b as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config a34626 and rejected config bec0fc as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config f72805 and rejected config a34626 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][smbo.py:225] A callback returned False. Abort is requested.
[INFO][smbo.py:333] Shutting down because the stop flag was set.
[INFO][abstract_initial_design.py:147] Using 10 initial design configurations and 0 additional configurations.
[INFO][smbo.py:498] Continuing from previous run.
[INFO][abstract_intensifier.py:288] Added existing seed 209652396 from runhistory to the intensifier.
[INFO][abstract_intensifier.py:306] Using only one seed for deterministic scenario.
[INFO][abstract_intensifier.py:595] Added config f47107 and rejected config f72805 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 988484 and rejected config f47107 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 9f53ac and rejected config 988484 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 204797 and rejected config 9f53ac as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config a0cecd and rejected config 204797 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 14f964 and rejected config a0cecd as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config df341a and rejected config 14f964 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config b9e9df and rejected config df341a as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 324506 and rejected config b9e9df as incumbent because it is not better than the incumbents on 1 instances:
[INFO][abstract_intensifier.py:595] Added config 3680e8 and rejected config 324506 as incumbent because it is not better than the incumbents on 1 instances:
[INFO][smbo.py:320] Finished 50 trials.
[INFO][smbo.py:328] Configuration budget is exhausted:
[INFO][smbo.py:329] --- Remaining wallclock time: inf
[INFO][smbo.py:330] --- Remaining cpu time: inf
[INFO][smbo.py:331] --- Remaining trials: 0
[INFO][abstract_intensifier.py:306] Using only one seed for deterministic scenario.
Default cost: 25.0
Incumbent cost of first run: 0.09616130553975616
[INFO][abstract_intensifier.py:306] Using only one seed for deterministic scenario.
Incumbent cost of continued run: 0.00016054829629549825

from __future__ import annotations

from ConfigSpace import Configuration, ConfigurationSpace, Float

from smac import Callback
from smac import HyperparameterOptimizationFacade as HPOFacade
from smac import Scenario
from smac.main.smbo import SMBO
from smac.runhistory import TrialInfo, TrialValue

__copyright__ = "Copyright 2021, AutoML.org Freiburg-Hannover"
__license__ = "3-clause BSD"


class StopCallback(Callback):
    def __init__(self, stop_after: int):
        self._stop_after = stop_after

    def on_tell_end(self, smbo: SMBO, info: TrialInfo, value: TrialValue) -> bool | None:
        """Called after the stats are updated and the trial is added to the runhistory. Optionally, returns false
        to gracefully stop the optimization.
        """
        if smbo.runhistory.finished == self._stop_after:
            return False

        return None


class QuadraticFunction:
    @property
    def configspace(self) -> ConfigurationSpace:
        cs = ConfigurationSpace(seed=0)
        x = Float("x", (-5, 5), default=-5)
        cs.add([x])

        return cs

    def train(self, config: Configuration, seed: int = 0) -> float:
        """Returns the y value of a quadratic function with a minimum at x=0."""
        x = config["x"]
        return x * x


if __name__ == "__main__":
    model = QuadraticFunction()

    # Scenario object specifying the optimization "environment"
    scenario = Scenario(model.configspace, deterministic=True, n_trials=50)
    stop_after = 10

    # Now we use SMAC to find the best hyperparameters
    smac = HPOFacade(
        scenario,
        model.train,  # We pass the target function here
        callbacks=[StopCallback(stop_after=stop_after)],
        overwrite=True,  # Overrides any previous results that are found that are inconsistent with the meta-data
    )

    incumbent = smac.optimize()
    assert smac.runhistory.finished == stop_after

    # Now, we want to continue the optimization
    # Make sure, we don't overwrite the last run
    smac2 = HPOFacade(
        scenario,
        model.train,
        overwrite=False,
    )

    # Check whether we get the same incumbent
    assert smac.intensifier.get_incumbent() == smac2.intensifier.get_incumbent()
    assert smac2.runhistory.finished == stop_after

    # And now we finish the optimization
    incumbent2 = smac2.optimize()

    default_cost = smac.validate(model.configspace.get_default_configuration())
    print(f"Default cost: {default_cost}")

    incumbent_cost = smac.validate(incumbent)
    print(f"Incumbent cost of first run: {incumbent_cost}")

    incumbent_cost = smac2.validate(incumbent2)
    print(f"Incumbent cost of continued run: {incumbent_cost}")

previous

Custom Callback

next

User Priors over the Optimum