DeepHyper: Distributed Neural Architecture and Hyperparameter Optimization for Machine Learning#

DeepHyper is a powerful Python package for automating machine learning tasks, particularly focused on optimizing hyperparameters, searching for optimal neural architectures, and quantifying uncertainty through the deep ensembles. With DeepHyper, users can easily perform these tasks on a single machine or distributed across multiple machines, making it ideal for use in a variety of environments. Whether you’re a beginner looking to optimize your machine learning models or an experienced data scientist looking to streamline your workflow, DeepHyper has something to offer. So why wait? Start using DeepHyper today and take your machine learning skills to the next level!

DeepHyper is specialized for machine learning tasks but it can also be used for generic black-box and gray-box optimization problems of expensive functions.

DeepHyper’s software architecture is designed to be modular and extensible. It is built on top of the following main sub-packages:

  • deephyper.ensemble: Tools to build ensembles of neural networks with uncertainty quantification.

  • deephyper.nas: Tools to define neural architecture search space and evaluation strategy.

  • deephyper.problem: Tools for defining neural architecture and hyper-parameter search problems.

  • deephyper.evaluator : Tools to distribute the evaluation of tasks (e.g., neural network trainings).

  • deephyper.search: Tools to define search strategies for neural architecture search and hyper-parameter optimization.

  • deephyper.stopper: Tools to define multi-fidelity strategies for neural architecture and hyper-parameter optimization.

DeepHyper installation requires Python >= 3.7.

Install instructions#

Install with pip

# For the most basic set of features (hyperparameter search)
pip install deephyper

# For the default set of features including:
# - hyperparameter search with transfer-learning
# - neural architecture search
# - deep ensembles
# - Ray-based distributed computing
# - Learning-curve extrapolation for multi-fidelity hyperparameter search
pip install "deephyper[default]"

More details about the installation process can be found at DeepHyper Installations.

Quick Start#

Open In Colab

The black-box function named run is defined by taking an input dictionnary named config which contains the different variables to optimize. Then the run-function is binded to an Evaluator in charge of distributing the computation of multiple evaluations. Finally, a Bayesian search named CBO is created and executed to find the values of config which MAXIMIZE the return value of run(config).

def run(job):
    # The suggested parameters are accessible in job.parameters (dict)
    x = job.parameters["x"]
    b = job.parameters["b"]

    if job.parameters["function"] == "linear":
        y = x + b
    elif job.parameters["function"] == "cubic":
        y = x**3 + b

    # Maximization!
    return y


# Necessary IF statement otherwise it will enter in a infinite loop
# when loading the 'run' function from a new process
if __name__ == "__main__":
    from deephyper.problem import HpProblem
    from deephyper.search.hps import CBO
    from deephyper.evaluator import Evaluator

    # define the variable you want to optimize
    problem = HpProblem()
    problem.add_hyperparameter((-10.0, 10.0), "x") # real parameter
    problem.add_hyperparameter((0, 10), "b") # discrete parameter
    problem.add_hyperparameter(["linear", "cubic"], "function") # categorical parameter

    # define the evaluator to distribute the computation
    evaluator = Evaluator.create(
        run,
        method="process",
        method_kwargs={
            "num_workers": 2,
        },
    )

    # define your search and execute it
    search = CBO(problem, evaluator, random_state=42)

    results = search.search(max_evals=100)
    print(results)

Which outputs the following results where the best parameters are with function == "cubic", x == 9.99 and b == 10.

    p:b p:function       p:x    objective  job_id  m:timestamp_submit  m:timestamp_gather
0     7     linear  8.831019    15.831019       1            0.064874            1.430992
1     4     linear  9.788889    13.788889       0            0.064862            1.453012
2     0      cubic  2.144989     9.869049       2            1.452692            1.468436
3     9     linear -9.236860    -0.236860       3            1.468123            1.483654
4     2      cubic -9.783865  -934.550818       4            1.483340            1.588162
..  ...        ...       ...          ...     ...                 ...                 ...
95    6      cubic  9.862098   965.197192      95           13.538506           13.671872
96   10      cubic  9.997512  1009.253866      96           13.671596           13.884530
97    6      cubic  9.965615   995.719961      97           13.884188           14.020144
98    5      cubic  9.998324  1004.497422      98           14.019737           14.154467
99    9      cubic  9.995800  1007.740379      99           14.154169           14.289366

The code defines a function run that takes a RunningJob job as input and returns the maximized objective y. The if block at the end of the code defines a black-box optimization process using the CBO (Centralized Bayesian Optimization) algorithm from the deephyper library.

The optimization process is defined as follows:

  1. A hyperparameter optimization problem is created using the HpProblem class from deephyper. In this case, the problem has a three variables. The x hyperparameter is a real variable in a range from -10.0 to 10.0. The b hyperparameter is a discrete variable in a range from 0 to 10. The function hyperparameter is a categorical variable with two possible values.

  2. An evaluator is created using the Evaluator.create method. The evaluator will be used to evaluate the function run with different configurations of suggested hyperparameters in the optimization problem. The evaluator uses the process method to distribute the evaluations across multiple worker processes, in this case 2 worker processes.

  3. A search object is created using the CBO class, the problem and evaluator defined earlier. The CBO algorithm is a derivative-free optimization method that uses a Bayesian optimization approach to explore the hyperparameter space.

  4. The optimization process is executed by calling the search.search method, which performs the evaluations of the run function with different configurations of the hyperparameters until a maximum number of evaluations (100 in this case) is reached.

  5. The results of the optimization process, including the optimal configuration of the hyperparameters and the corresponding objective value, are printed to the console.

Warning

All search algorithms are MAXIMIZING the objective function. If you want to MINIMIZE the objective function, you have to return the negative of you objective.

Table of Contents#

API Reference

Indices and tables#