AutoSampler, error when switching to CMAES sampler

[rodrigoscoelho]

Expected behavior

When using AutoSampler, it starts as expected with the first 250 trials, and, when trying to switch to CMAES sampler, as expected when the numer of trials > 250, it returns:
ERROR - Error in _suggest_params: No compatible source_trials

Environment

• The package of OptunaHub which you are using:
• Optuna version: 4.1
• OptunaHub version:
• Python version:
• OS:
• (Optional) Other libraries and their versions:

Error messages, stack traces, or logs

ERROR - Error in _suggest_params: No compatible source_trials

Steps to reproduce

1. Run a optimization using autosampler. single objective, no constraints func.
2. wait until trial 250

Additional context (optional)

No response

[nabenabe0928]

@rodrigoscoelho
Thank you for reporting this issue!
Could you possibly give me the following?

1. Versions: cmaes
2. The code you used

[nabenabe0928]

The following code did not reproduce your issue.

import optuna
import optunahub


def objective(trial: optuna.Trial) -> float:
    x = trial.suggest_float("x", -5, 5)
    y = trial.suggest_float("y", -5, 5)
    return x**2 + y**2


sampler = optunahub.load_module("samplers/auto_sampler").AutoSampler()
study = optuna.create_study(sampler=sampler)
study.optimize(objective, n_trials=300)

optuna==4.1.0
optunahub==0.1.0
cmaes==0.11.1

[rodrigoscoelho]

Maybe this is related to the fact that my first trial is outside the dimensions of the others, as I always use a manual suggestion for the initial attempt. It’s a signal controller optimizer that uses the Ask() and Tell() methodology for optimization. However, if the GPSampler accepts this trial, the other samplers should also accept it. Perhaps the issue lies in the CMAES sampler when applying the filter to prepare the warmup. Em seg., 2 de dez. de 2024 às 12:43, Shuhei Watanabe < ***@***.***> escreveu:

…

The following code did not reproduce your issue. import optunaimport optunahub def objective(trial: optuna.Trial) -> float: x = trial.suggest_float("x", -5, 5) y = trial.suggest_float("y", -5, 5) return x**2 + y**2 sampler = optunahub.load_module("samplers/auto_sampler").AutoSampler()study = optuna.create_study(sampler=sampler)study.optimize(objective, n_trials=300) — Reply to this email directly, view it on GitHub <#194 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABUXTGU74W4PHDMBISNNXLD2DR52HAVCNFSM6AAAAABS3UI5J6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJRHA4DMNZQGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[nabenabe0928]

Then could you, @rodrigoscoelho, please tell me the distribution given to ask and the search_space in the objective?

[rodrigoscoelho]

I can show you the table of parameters. If you look, the only trial using cycle 90 is the trial 0. The others are cycle = 120 (max and min = 120). [image: image.png] Em seg., 2 de dez. de 2024 às 15:52, Shuhei Watanabe < ***@***.***> escreveu:

…

Then could you, @rodrigoscoelho <https://github.com/rodrigoscoelho>, please tell me the distribution given to ask and the search_space in the objective? — Reply to this email directly, view it on GitHub <#194 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABUXTGQON7PREW56YIQIEWL2DST5NAVCNFSM6AAAAABS3UI5J6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJSGQZTINJSHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[rodrigoscoelho]

Here goes the image

[nabenabe0928]

@rodrigoscoelho

I suspect the search space you feeded via ask method was incompatible.
Could you give me a simpler code that reproduces your issue?

• https://github.com/optuna/optuna/blob/master/optuna/samplers/_cmaes.py#L642-L649
• https://github.com/optuna/optuna/blob/master/optuna/samplers/_cmaes.py#L770-L774