Skip to main content

Scalable asynchronous neural architecture and hyperparameter search for deep neural networks.

Project description

DOI GitHub tag (latest by date) Documentation Status PyPI - License PyPI - Downloads Open In Colab

What is DeepHyper?

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 use of 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!

Install Instructions

Installation 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.

Quickstart

Open In Colab

The black-box function named run is defined by taking an input job named job which contains the different variables to optimize job.parameters. Then the run-function is bound 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(job).

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.hpo import HpProblem
    from deephyper.hpo 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 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.

How do I learn more?

Contributions

Find the list of contributors on the DeepHyper Authors page of the Documentation.

Citing DeepHyper

If you wish to cite the Software, please use the following:

@misc{deephyper_software,
    title = {"DeepHyper: A Python Package for Scalable Neural Architecture and Hyperparameter Search"},
    author = {Balaprakash, Prasanna and Egele, Romain and Salim, Misha and Maulik, Romit and Vishwanath, Venkat and Wild, Stefan and others},
    organization = {DeepHyper Team},
    year = 2018,
    url = {https://github.com/deephyper/deephyper}
} 

Find all our publications on the Research & Publication page of the Documentation.

How can I participate?

Questions, comments, feature requests, bug reports, etc. can be directed to:

  • Issues on GitHub

Patches through pull requests are much appreciated on the software itself as well as documentation. Optionally, please include in your first patch a credit for yourself in the list above.

The DeepHyper Team uses git-flow to organize the development: Git-Flow cheatsheet. For tests we are using: Pytest.

Acknowledgments

  • Scalable Data-Efficient Learning for Scientific Domains, U.S. Department of Energy 2018 Early Career Award funded by the Advanced Scientific Computing Research program within the DOE Office of Science (2018--Present)
  • Argonne Leadership Computing Facility: This research used resources of the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357.
  • SLIK-D: Scalable Machine Learning Infrastructures for Knowledge Discovery, Argonne Computing, Environment and Life Sciences (CELS) Laboratory Directed Research and Development (LDRD) Program (2016--2018)

Copyright and license

Copyright © 2019, UChicago Argonne, LLC

DeepHyper is distributed under the terms of BSD License. See LICENSE

Argonne Patent & Intellectual Property File Number: SF-19-007

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

deephyper-0.8.1.tar.gz (201.5 kB view hashes)

Uploaded Source

Built Distribution

deephyper-0.8.1-py2.py3-none-any.whl (271.4 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page