Optimizing average cross-validation performance with BOHB

An example for the usage of Hyperband intensifier in SMAC with multiple instances. We optimize a SGD classifier on the digits dataset as multiple binary classification problems using “Hyperband” intensification. We split the digits dataset (10 classes) into 45 binary datasets.

In this example, we use instances as the budget in hyperband and optimize the average cross validation accuracy. An “Instance” represents a specific scenario/condition (eg: different datasets, subsets, transformations) for the algorithm to run. SMAC then returns the algorithm that had the best performance across all the instances. In this case, an instance is a binary dataset i.e., digit-2 vs digit-3.

import itertools
import logging
import warnings

import numpy as np
from ConfigSpace.hyperparameters import CategoricalHyperparameter, UniformFloatHyperparameter
from sklearn import datasets
from sklearn.exceptions import ConvergenceWarning
from sklearn.linear_model import SGDClassifier
from sklearn.model_selection import cross_val_score, StratifiedKFold

# Import ConfigSpace and different types of parameters
from smac.configspace import ConfigurationSpace
from smac.facade.smac_bohb_facade import BOHB4HPO
# Import SMAC-utilities
from smac.scenario.scenario import Scenario

# We load the MNIST-dataset (a widely used benchmark) and split it into a collection of binary datasets
digits = datasets.load_digits()
instances = [[str(a) + str(b)] for a, b in itertools.combinations(digits.target_names, 2)]

def generate_instances(a: int, b: int):
    Function to select data for binary classification from the digits dataset
    a & b are the two classes
    # get indices of both classes
    indices = np.where(np.logical_or(a == digits.target, b == digits.target))
    # get data
    data = digits.data[indices]
    target = digits.target[indices]
    return data, target

# Target Algorithm
def sgd_from_cfg(cfg, seed, instance):
    """ Creates a SGD classifier based on a configuration and evaluates it on the
    digits dataset using cross-validation.

    cfg: Configuration (ConfigSpace.ConfigurationSpace.Configuration)
        Configuration containing the parameters.
        Configurations are indexable!
    seed: int or RandomState
        used to initialize the svm's random generator
    instance: str
        used to represent the instance to use (the 2 classes to consider in this case)

        A crossvalidated mean score for the SGD classifier on the loaded data-set.

    with warnings.catch_warnings():
        warnings.filterwarnings('ignore', category=ConvergenceWarning)

        # SGD classifier using given configuration
        clf = SGDClassifier(loss='log', penalty='elasticnet', alpha=cfg['alpha'], l1_ratio=cfg['l1_ratio'],
                            learning_rate=cfg['learning_rate'], eta0=cfg['eta0'],
                            max_iter=30, early_stopping=True, random_state=seed)

        # get instance
        data, target = generate_instances(int(instance[0]), int(instance[1]))

        cv = StratifiedKFold(n_splits=4, random_state=seed, shuffle=True)  # to make CV splits consistent
        scores = cross_val_score(clf, data, target, cv=cv)
    return 1 - np.mean(scores)  # Minimize!

logger = logging.getLogger("Hyperband-instances-example")
logging.basicConfig(level=logging.INFO)  # logging.DEBUG for debug output

# Build Configuration Space which defines all parameters and their ranges
cs = ConfigurationSpace()

# We define a few possible parameters for the SGD classifier
alpha = UniformFloatHyperparameter("alpha", 0, 1, default_value=1.0)
l1_ratio = UniformFloatHyperparameter("l1_ratio", 0, 1, default_value=0.5)
learning_rate = CategoricalHyperparameter("learning_rate", choices=['constant', 'invscaling', 'adaptive'],
eta0 = UniformFloatHyperparameter("eta0", 0.00001, 1, default_value=0.1, log=True)
# Add the parameters to configuration space
cs.add_hyperparameters([alpha, l1_ratio, learning_rate, eta0])

# SMAC scenario object
scenario = Scenario({"run_obj": "quality",  # we optimize quality (alternative to runtime)
                     "wallclock-limit": 100,  # max duration to run the optimization (in seconds)
                     "cs": cs,  # configuration space
                     "deterministic": True,
                     "limit_resources": True,  # Uses pynisher to limit memory and runtime
                     "memory_limit": 3072,  # adapt this to reasonable value for your hardware
                     "cutoff": 3,  # runtime limit for the target algorithm
                     "instances": instances  # Optimize across all given instances

# intensifier parameters
# if no argument provided for budgets, hyperband decides them based on the number of instances available
intensifier_kwargs = {'initial_budget': 1, 'max_budget': 45, 'eta': 3,
                      'instance_order': None,  # You can also shuffle the order of using instances by this parameter.
                      # 'shuffle' will shuffle instances before each SH run and
                      # 'shuffle_once' will shuffle instances once before the 1st SH iteration begins

# To optimize, we pass the function to the SMAC-object
smac = BOHB4HPO(scenario=scenario, rng=np.random.RandomState(42),
                intensifier_kwargs=intensifier_kwargs)  # all arguments related to intensifier can be passed like this

# Example call of the function
# It returns: Status, Cost, Runtime, Additional Infos
def_costs = []
for i in instances:
    cost = smac.get_tae_runner().run(cs.get_default_configuration(), i[0])[1]
print("Value for default configuration: %.4f" % (np.mean(def_costs)))

# Start optimization
    incumbent = smac.optimize()
    incumbent = smac.solver.incumbent

inc_costs = []
for i in instances:
    cost = smac.get_tae_runner().run(incumbent, i[0])[1]
print("Optimized Value: %.4f" % (np.mean(inc_costs)))

Total running time of the script: ( 0 minutes 0.000 seconds)

Gallery generated by Sphinx-Gallery