Document Behavior of `optuna.pruners.PercentilePruner`, `optuna.pruners.SuccessiveHalvingPruner` and `optuna.pruners.HyperbandPruner` #6092

ParagEkbote · 2025-05-25T19:15:30Z

Motivation

I have added documentation of how the Percentile Pruner, SuccessiveHalvingPruner and HyperbandPruner handles Nan values.

It is important to note that we are using synthetic dataset for testing.

Could you please review?

cc: @HideakiImamura

Description of the changes

I have used the following scripts to test it out:

Percentile Pruner

import optuna
import numpy as np
import logging
import sys
from optuna.pruners import SuccessiveHalvingPruner

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

def nan_objective(trial):
    """An objective function that returns NaN for every 3rd trial."""
    trial_number = trial.number
    
    # Every 3rd trial returns NaN
    if trial_number % 3 == 0:
        logger.info(f"Trial {trial_number}: Returning NaN")
        return float('nan')
    else:
        value = trial.suggest_float('x', -10, 10)
        result = value ** 2  # Simple quadratic function
        logger.info(f"Trial {trial_number}: Returning {result}")
        return result

def partial_nan_objective(trial):
    """An objective function where some intermediate values are NaN,
    specifically designed to test SuccessiveHalvingPruner."""
    x = trial.suggest_float('x', -10, 10)
    n_steps = 10
    
    # Report intermediate values
    for step in range(n_steps):
        # Calculate a result (simple quadratic plus step index)
        raw_value = x ** 2 + step
        
        # Determine if this step should report NaN based on different patterns
        report_nan = False
        
        # Pattern 1: Every 3rd step is NaN
        if step % 3 == 0:
            report_nan = True
            
        # Pattern 2: For trials divisible by 5, make later steps NaN
        if trial.number % 5 == 0 and step > 5:
            report_nan = True
            
        # Report value
        if report_nan:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
            trial.report(float('nan'), step)
        else:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {raw_value}")
            trial.report(raw_value, step)
        
        # Check if the trial should be pruned
        if trial.should_prune():
            logger.info(f"Trial {trial.number} pruned at step {step}")
            raise optuna.exceptions.TrialPruned()
    
    return x ** 2

def pruning_pattern_objective(trial):
    """Objective function with specific NaN patterns at pruning-relevant steps."""
    x = trial.suggest_float('x', -10, 10)
    
    # These are the reduction factors we'll test
    reduction_factors = [3, 4]
    
    for rf in reduction_factors:
    
        max_step = rf ** 3  # Go up to rf^3 steps
        steps = []
        val = 0
        while val <= max_step:
            steps.append(val)
            if val == 0:
                val = 1
            else:
                val *= rf
        
        # Run a mini-test with this reduction factor
        logger.info(f"\n--- Testing with reduction_factor={rf} ---")
        study = optuna.create_study(
            pruner=SuccessiveHalvingPruner(
                min_resource=1,
                reduction_factor=rf,
                min_early_stopping_rate=0
            ),
            direction="minimize"
        )
        
        # Create a custom objective for this reduction factor
        def rf_objective(trial):
            value = trial.suggest_float('x', -10, 10)
            result = value ** 2
            
            for i, step in enumerate(steps):
                # For some trials, introduce NaNs at specific points
                if trial.number % 4 == 0 and i % 2 == 0:  # Every 4th trial, every other checkpoint
                    logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
                    trial.report(float('nan'), step)
                else:
                    # Add some noise to the result based on step
                    step_result = result + step * 0.1
                    logger.info(f"Trial {trial.number}, Step {step}: Reporting {step_result}")
                    trial.report(step_result, step)
                
                if trial.should_prune():
                    logger.info(f"Trial {trial.number} pruned at step {step}")
                    raise optuna.exceptions.TrialPruned()
            
            return result
        
        study.optimize(rf_objective, n_trials=12)
        
        # Print results
        logger.info(f"Results for reduction_factor={rf}:")
        for trial in study.trials:
            logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
            if trial.intermediate_values:
                logger.info(f"  Intermediate values: {trial.intermediate_values}")

def test_successive_halving_pruner_with_nan_values():
    """Test how SuccessiveHalvingPruner handles NaN values in different configurations."""
    
    # Test 1: Basic study with NaN return values
    logger.info("\n=== Test 1: Basic study with NaN final values ===")
    study1 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=2, min_early_stopping_rate=0),
        direction="minimize"
    )
    study1.optimize(nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials:")
    for trial in study1.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
    
    # Test 2: Study with NaN intermediate values
    logger.info("\n=== Test 2: Study with NaN intermediate values ===")
    study2 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=2, min_early_stopping_rate=0),
        direction="minimize"
    )
    study2.optimize(partial_nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials with intermediate values:")
    for trial in study2.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")
    
    # Test 3: Study with higher reduction factor
    logger.info("\n=== Test 3: Study with higher reduction factor ===")
    study3 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=4, min_early_stopping_rate=0),
        direction="minimize"
    )
    study3.optimize(partial_nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials with higher reduction factor:")
    for trial in study3.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")
    
    # Test 4: Test specific pruning patterns with NaN values
    logger.info("\n=== Test 4: Testing specific pruning patterns with NaN values ===")
    pruning_pattern_objective(trial)  # No trial needed here as we create studies inside

if __name__ == "__main__":
    test_successive_halving_pruner_with_nan_values()

SuccessiveHalvingPruner

import optuna
import numpy as np
import logging
import sys
from optuna.pruners import SuccessiveHalvingPruner

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

def nan_objective(trial):
    """An objective function that returns NaN for every 3rd trial."""
    trial_number = trial.number
    
    # Every 3rd trial returns NaN
    if trial_number % 3 == 0:
        logger.info(f"Trial {trial_number}: Returning NaN")
        return float('nan')
    else:
        value = trial.suggest_float('x', -10, 10)
        result = value ** 2  # Simple quadratic function
        logger.info(f"Trial {trial_number}: Returning {result}")
        return result

def partial_nan_objective(trial):
    """An objective function where some intermediate values are NaN,
    specifically designed to test SuccessiveHalvingPruner."""
    x = trial.suggest_float('x', -10, 10)
    n_steps = 10
    
    # Report intermediate values
    for step in range(n_steps):
        # Calculate a result (simple quadratic plus step index)
        raw_value = x ** 2 + step
        
        # Determine if this step should report NaN based on different patterns
        report_nan = False
        
        # Pattern 1: Every 3rd step is NaN
        if step % 3 == 0:
            report_nan = True
            
        # Pattern 2: For trials divisible by 5, make later steps NaN
        if trial.number % 5 == 0 and step > 5:
            report_nan = True
            
        # Report value
        if report_nan:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
            trial.report(float('nan'), step)
        else:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {raw_value}")
            trial.report(raw_value, step)
        
        # Check if the trial should be pruned
        if trial.should_prune():
            logger.info(f"Trial {trial.number} pruned at step {step}")
            raise optuna.exceptions.TrialPruned()
    
    return x ** 2

def pruning_pattern_objective(trial):
    """Objective function with specific NaN patterns at pruning-relevant steps."""
    x = trial.suggest_float('x', -10, 10)
    
    # These are the reduction factors we'll test
    reduction_factors = [3, 4]
    
    for rf in reduction_factors:
        # Create steps that align with SuccessiveHalvingPruner's expected checkpoints
        # For reduction_factor=3, these would be [0, 1, 3, 9, 27, ...]
        # For reduction_factor=4, these would be [0, 1, 4, 16, 64, ...]
        max_step = rf ** 3  # Go up to rf^3 steps
        steps = []
        val = 0
        while val <= max_step:
            steps.append(val)
            if val == 0:
                val = 1
            else:
                val *= rf
        
        # Run a mini-test with this reduction factor
        logger.info(f"\n--- Testing with reduction_factor={rf} ---")
        study = optuna.create_study(
            pruner=SuccessiveHalvingPruner(
                min_resource=1,
                reduction_factor=rf,
                min_early_stopping_rate=0
            ),
            direction="minimize"
        )
        
        # Create a custom objective for this reduction factor
        def rf_objective(trial):
            value = trial.suggest_float('x', -10, 10)
            result = value ** 2
            
            for i, step in enumerate(steps):
                # For some trials, introduce NaNs at specific points
                if trial.number % 4 == 0 and i % 2 == 0:  # Every 4th trial, every other checkpoint
                    logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
                    trial.report(float('nan'), step)
                else:
                    # Add some noise to the result based on step
                    step_result = result + step * 0.1
                    logger.info(f"Trial {trial.number}, Step {step}: Reporting {step_result}")
                    trial.report(step_result, step)
                
                if trial.should_prune():
                    logger.info(f"Trial {trial.number} pruned at step {step}")
                    raise optuna.exceptions.TrialPruned()
            
            return result
        
        study.optimize(rf_objective, n_trials=12)
        
        # Print results
        logger.info(f"Results for reduction_factor={rf}:")
        for trial in study.trials:
            logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
            if trial.intermediate_values:
                logger.info(f"  Intermediate values: {trial.intermediate_values}")

def test_successive_halving_pruner_with_nan_values():
    """Test how SuccessiveHalvingPruner handles NaN values in different configurations."""
    
    # Test 1: Basic study with NaN return values
    logger.info("\n=== Test 1: Basic study with NaN final values ===")
    study1 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=2, min_early_stopping_rate=0),
        direction="minimize"
    )
    study1.optimize(nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials:")
    for trial in study1.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
    
    # Test 2: Study with NaN intermediate values
    logger.info("\n=== Test 2: Study with NaN intermediate values ===")
    study2 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=2, min_early_stopping_rate=0),
        direction="minimize"
    )
    study2.optimize(partial_nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials with intermediate values:")
    for trial in study2.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")
    
    # Test 3: Study with higher reduction factor
    logger.info("\n=== Test 3: Study with higher reduction factor ===")
    study3 = optuna.create_study(
        pruner=SuccessiveHalvingPruner(min_resource=1, reduction_factor=4, min_early_stopping_rate=0),
        direction="minimize"
    )
    study3.optimize(partial_nan_objective, n_trials=10)
    
    # Print results
    logger.info("Completed trials with higher reduction factor:")
    for trial in study3.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")
    
    # Test 4: Test specific pruning patterns with NaN values
    logger.info("\n=== Test 4: Testing specific pruning patterns with NaN values ===")
    pruning_pattern_objective(trial)  # No trial needed here as we create studies inside

if __name__ == "__main__":
    test_successive_halving_pruner_with_nan_values()

HyperbandPruner

import optuna
import numpy as np
import logging

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

def test_hyperband_with_nan_final_values():
    """Test 1: Basic study with NaN final values"""
    logger.info("\n=== Test 1: HyperbandPruner with NaN final values ===")
    
    def objective(trial):
        x = trial.suggest_float("x", -10, 10)
        # Return NaN for specific trials
        if trial.number % 4 == 0:
            logger.info(f"Trial {trial.number}: Returning NaN")
            return float('nan')
        
        value = (x - 0)**2
        logger.info(f"Trial {trial.number}: Returning {value}")
        return value
    
    study = optuna.create_study(
        pruner=optuna.pruners.HyperbandPruner(
            min_resource=1,
            max_resource=10,
            reduction_factor=3
        ),
        direction="minimize"
    )
    
    try:
        study.optimize(objective, n_trials=10)
    except Exception as e:
        logger.error(f"Optimization failed: {e}")
    
    logger.info("Completed trials:")
    for trial in study.trials:
        logger.info(f"Trial {trial.number}: State={trial.state.value}, Value={trial.value}")

def test_hyperband_with_nan_intermediate_values():
    """Test 2: Study with NaN intermediate values"""
    logger.info("\n=== Test 2: HyperbandPruner with NaN intermediate values ===")
    
    def objective(trial):
        x = trial.suggest_float("x", -10, 10)
        final_value = (x - 0)**2
        
        for step in range(10):
            # Report NaN at specific steps
            if step % 3 == 0:
                value_to_report = float('nan')
            else:
                value_to_report = final_value + step
            
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {value_to_report}")
            trial.report(value_to_report, step)
            
            if trial.should_prune():
                logger.info(f"Trial {trial.number} pruned at step {step}")
                raise optuna.TrialPruned()
        
        return final_value
    
    study = optuna.create_study(
        pruner=optuna.pruners.HyperbandPruner(
            min_resource=1,
            max_resource=10,
            reduction_factor=3
        ),
        direction="minimize"
    )
    
    try:
        study.optimize(objective, n_trials=10)
    except Exception as e:
        logger.error(f"Optimization failed: {e}")
    
    logger.info("Completed trials with intermediate values:")
    for trial in study.trials:
        logger.info(f"Trial {trial.number}: State={trial.state.value}, Value={trial.value}")
        logger.info(f"  Intermediate values: {trial.intermediate_values}")

if __name__ == "__main__":
    test_hyperband_with_nan_final_values()
    test_hyperband_with_nan_intermediate_values()

nabenabe0928 · 2025-05-26T06:56:15Z

To reviewers, please note this PR:

Document Behavior of optuna.pruners.MedianPruner and optuna.pruners.PatientPruner #6055

At first glance, the changes in this PR do not follow the pattern in the previous PR.

HideakiImamura · 2025-05-28T03:10:46Z

@gen740 @kAIto47802 @sawa3030 Could you review this PR?

sawa3030 · 2025-05-30T06:34:26Z

As @nabenabe0928 pointed out, I think it would be better to follow the approach in #6055 by at least addressing the following two points: what would happen if all intermediate values are NaN , and how typical NaN values are handled.

nabenabe0928 · 2025-06-06T04:25:00Z

@ParagEkbote
Are you still down for the update?

ParagEkbote · 2025-06-06T09:35:12Z

I apologize for not updating the PR. I was a bit busy and in the following week, I will update the PR as suggested.

cc: @nabenabe0928

ParagEkbote · 2025-06-12T09:38:04Z

I'd like to thank the reviewers for their patience :)

I have updated the docs as per the previous PR, the updated scripts are:

Percentile Pruner:

import optuna
import numpy as np
import logging
import sys

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

def nan_objective(trial):
    trial_number = trial.number
    if trial_number % 3 == 0:
        logger.info(f"Trial {trial_number}: Returning NaN")
        return float('nan')
    else:
        value = trial.suggest_float('x', -10, 10)
        result = value ** 2
        logger.info(f"Trial {trial_number}: Returning {result}")
        return result

def partial_nan_objective(trial):
    x = trial.suggest_float('x', -10, 10)
    for step in range(10):
        value = x ** 2 + step
        if step % 3 == 0:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
            trial.report(float('nan'), step)
        else:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {value}")
            trial.report(value, step)
        if trial.should_prune():
            logger.info(f"Trial {trial.number} pruned at step {step}")
            raise optuna.exceptions.TrialPruned()
    return x ** 2

def test_percentile_pruner_with_nan_values():
    logger.info("\n=== Test 1: Basic study with NaN final values ===")
    study1 = optuna.create_study(
        pruner=optuna.pruners.PercentilePruner(percentile=25.0, n_startup_trials=0, n_warmup_steps=0),
        direction="minimize"
    )
    study1.optimize(nan_objective, n_trials=10)

    logger.info("Completed trials:")
    for trial in study1.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")

    logger.info("\n=== Test 2: Study with NaN intermediate values ===")
    study2 = optuna.create_study(
        pruner=optuna.pruners.PercentilePruner(percentile=25.0, n_startup_trials=5, n_warmup_steps=2),
        direction="minimize"
    )
    study2.optimize(partial_nan_objective, n_trials=10)

    logger.info("Completed trials with intermediate values:")
    for trial in study2.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")

if __name__ == "__main__":
    test_percentile_pruner_with_nan_values()

SuccessiveHalving Pruner:

import optuna
import numpy as np
import logging

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

def nan_objective(trial):
    if trial.number % 3 == 0:
        logger.info(f"Trial {trial.number}: Returning NaN")
        return float('nan')
    value = trial.suggest_float('x', -10, 10)
    result = value ** 2
    logger.info(f"Trial {trial.number}: Returning {result}")
    return result

def partial_nan_objective(trial):
    x = trial.suggest_float('x', -10, 10)
    for step in range(10):
        value = x ** 2 + step
        if step % 3 == 0:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
            trial.report(float('nan'), step)
        else:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {value}")
            trial.report(value, step)
        if trial.should_prune():
            logger.info(f"Trial {trial.number} pruned at step {step}")
            raise optuna.exceptions.TrialPruned()
    return x ** 2

def test_successive_halving_with_nan_values():
    logger.info("\n=== Test 1: SuccessiveHalving with NaN final values ===")
    study1 = optuna.create_study(
        pruner=optuna.pruners.SuccessiveHalvingPruner(min_resource=1, reduction_factor=2),
        direction="minimize"
    )
    study1.optimize(nan_objective, n_trials=10)

    logger.info("Completed trials:")
    for trial in study1.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")

    logger.info("\n=== Test 2: Intermediate NaNs with SuccessiveHalving ===")
    study2 = optuna.create_study(
        pruner=optuna.pruners.SuccessiveHalvingPruner(min_resource=1, reduction_factor=2),
        direction="minimize"
    )
    study2.optimize(partial_nan_objective, n_trials=10)

    logger.info("Completed trials:")
    for trial in study2.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")

if __name__ == "__main__":
    test_successive_halving_with_nan_values()

Hyperband Pruner:

import optuna
import numpy as np
import logging

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

def nan_objective(trial):
    if trial.number % 3 == 0:
        logger.info(f"Trial {trial.number}: Returning NaN")
        return float('nan')
    value = trial.suggest_float('x', -10, 10)
    result = value ** 2
    logger.info(f"Trial {trial.number}: Returning {result}")
    return result

def partial_nan_objective(trial):
    x = trial.suggest_float('x', -10, 10)
    for step in range(10):
        value = x ** 2 + step
        if step % 3 == 0:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting NaN")
            trial.report(float('nan'), step)
        else:
            logger.info(f"Trial {trial.number}, Step {step}: Reporting {value}")
            trial.report(value, step)
        if trial.should_prune():
            logger.info(f"Trial {trial.number} pruned at step {step}")
            raise optuna.exceptions.TrialPruned()
    return x ** 2

def test_hyperband_pruner_with_nan_values():
    logger.info("\n=== Test 1: Study with Hyperband and NaN final values ===")
    study1 = optuna.create_study(
        pruner=optuna.pruners.HyperbandPruner(),
        direction="minimize"
    )
    study1.optimize(nan_objective, n_trials=10)

    logger.info("Completed trials:")
    for trial in study1.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")

    logger.info("\n=== Test 2: Hyperband study with intermediate NaNs ===")
    study2 = optuna.create_study(
        pruner=optuna.pruners.HyperbandPruner(min_resource=1, max_resource=10, reduction_factor=3),
        direction="minimize"
    )
    study2.optimize(partial_nan_objective, n_trials=10)

    logger.info("Completed trials:")
    for trial in study2.trials:
        logger.info(f"Trial {trial.number}: State={trial.state}, Value={trial.value}")
        if trial.intermediate_values:
            logger.info(f"  Intermediate values: {trial.intermediate_values}")

if __name__ == "__main__":
    test_hyperband_pruner_with_nan_values()

Could you please review the changes?

gen740

I added the minor comment. PTAL!

optuna/pruners/_hyperband.py

optuna/pruners/_successive_halving.py

Co-authored-by: Gen <[email protected]>

codecov · 2025-06-13T05:36:52Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.42%. Comparing base (5ccdf30) to head (81cec3c).
Report is 23 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6092      +/-   ##
==========================================
+ Coverage   88.40%   88.42%   +0.02%     
==========================================
  Files         207      207              
  Lines       14029    14029              
==========================================
+ Hits        12402    12405       +3     
+ Misses       1627     1624       -3

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

sawa3030 · 2025-06-13T08:15:06Z

optuna/pruners/_successive_halving.py

@@ -17,6 +17,11 @@ class SuccessiveHalvingPruner(BasePruner):
    `Asynchronous Successive Halving <https://proceedings.mlsys.org/paper_files/paper/2020/file/
    a06f20b349c6cf09a6b171c71b88bbfc-Paper.pdf>`__ for detailed descriptions.

+    The pruner handles NaN values in the following manner:


According to the implementation, the pruning decision is based on the intermediate value at the last reported step, as well as the intermediate values of other trials at the same rung key. Could you update the explanation to reflect this behavior more accurately?

ParagEkbote added 2 commits May 25, 2025 18:57

update the docs on the pruning behaviour.

0c98423

Merge branch 'optuna:master' into Document-Nan-Bahviour

6ae8b04

HideakiImamura assigned gen740 and kAIto47802 May 28, 2025

y0z added the document Documentation related. label Jun 4, 2025

nabenabe0928 assigned nabenabe0928 and unassigned kAIto47802 Jun 6, 2025

ParagEkbote added 2 commits June 12, 2025 11:45

Merge branch 'optuna:master' into Document-Nan-Bahviour

c3a87cb

update the documentation about NaN values.

a282b42

gen740 reviewed Jun 13, 2025

View reviewed changes

optuna/pruners/_hyperband.py Outdated Show resolved Hide resolved

optuna/pruners/_successive_halving.py Outdated Show resolved Hide resolved

ParagEkbote and others added 2 commits June 13, 2025 10:51

Merge branch 'optuna:master' into Document-Nan-Bahviour

b438e5f

Apply suggestions from code review

81cec3c

Co-authored-by: Gen <[email protected]>

ParagEkbote requested a review from gen740 June 13, 2025 05:45

sawa3030 reviewed Jun 13, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Document Behavior of `optuna.pruners.PercentilePruner`, `optuna.pruners.SuccessiveHalvingPruner` and `optuna.pruners.HyperbandPruner` #6092

Document Behavior of `optuna.pruners.PercentilePruner`, `optuna.pruners.SuccessiveHalvingPruner` and `optuna.pruners.HyperbandPruner` #6092

Uh oh!

ParagEkbote commented May 25, 2025

Uh oh!

nabenabe0928 commented May 26, 2025

Uh oh!

HideakiImamura commented May 28, 2025

Uh oh!

sawa3030 commented May 30, 2025

Uh oh!

nabenabe0928 commented Jun 6, 2025

Uh oh!

ParagEkbote commented Jun 6, 2025

Uh oh!

ParagEkbote commented Jun 12, 2025

Uh oh!

gen740 left a comment

Uh oh!

Uh oh!

Uh oh!

codecov bot commented Jun 13, 2025

Uh oh!

sawa3030 Jun 13, 2025

Uh oh!

Uh oh!

Uh oh!

Document Behavior of optuna.pruners.PercentilePruner, optuna.pruners.SuccessiveHalvingPruner and optuna.pruners.HyperbandPruner #6092

Are you sure you want to change the base?

Document Behavior of optuna.pruners.PercentilePruner, optuna.pruners.SuccessiveHalvingPruner and optuna.pruners.HyperbandPruner #6092

Uh oh!

Conversation

ParagEkbote commented May 25, 2025

Motivation

Description of the changes

Uh oh!

nabenabe0928 commented May 26, 2025

Uh oh!

HideakiImamura commented May 28, 2025

Uh oh!

sawa3030 commented May 30, 2025

Uh oh!

nabenabe0928 commented Jun 6, 2025

Uh oh!

ParagEkbote commented Jun 6, 2025

Uh oh!

ParagEkbote commented Jun 12, 2025

Uh oh!

gen740 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

codecov bot commented Jun 13, 2025

Codecov Report

Uh oh!

sawa3030 Jun 13, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Document Behavior of `optuna.pruners.PercentilePruner`, `optuna.pruners.SuccessiveHalvingPruner` and `optuna.pruners.HyperbandPruner` #6092

Document Behavior of `optuna.pruners.PercentilePruner`, `optuna.pruners.SuccessiveHalvingPruner` and `optuna.pruners.HyperbandPruner` #6092