DeepHyper: Scalable Neural Architecture and Hyperparameter Search for Deep Neural Networks#


DeepHyper is a distributed machine learning (AutoML) package for automating the development of deep neural networks for scientific applications. It can run on a single laptop as well as on 1,000 of nodes.

It comprises different tools such as:

  • Optimizing hyper-parameters for a given black-box function.

  • Neural architecture search to discover high-performing deep neural network with variable operations and connections.

  • Automated machine learning, to easily experiment many learning algorithms from Scikit-Learn.

DeepHyper provides an infrastructure that targets experimental research in NAS and HPS methods, scalability, and portability across diverse supercomputers. It comprises three main modules:

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

  • deephyper.evaluator : A simple interface to dispatch model evaluation tasks. Implementations range from subprocess for laptop experiments to ray for large-scale runs on HPC systems.

  • Search methods for NAS and HPS. By extending the generic Search class, one can easily add new NAS or HPS methods to DeepHyper.

DeepHyper installation requires Python >= 3.7.

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(config: dict):
    return -config["x"]**2

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

    # define the variable you want to optimize
    problem = HpProblem()
    problem.add_hyperparameter((-10.0, 10.0), "x")

    # define the evaluator to distribute the computation
    evaluator = Evaluator.create(
            "num_workers": 2,

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

    results =

Which outputs the following where the best x found is clearly around 0.

        x  job_id     objective  timestamp_submit  timestamp_gather
0  -7.744105       1 -5.997117e+01          0.011047          0.037649
1  -9.058254       2 -8.205196e+01          0.011054          0.056398
2  -1.959750       3 -3.840621e+00          0.049750          0.073166
3  -5.150553       4 -2.652819e+01          0.065681          0.089355
4  -6.697095       5 -4.485108e+01          0.082465          0.158050
..       ...     ...           ...               ...               ...
95 -0.034096      96 -1.162566e-03         26.479630         26.795639
96 -0.034204      97 -1.169901e-03         26.789255         27.155481
97 -0.037873      98 -1.434366e-03         27.148506         27.466934
98 -0.000073      99 -5.387088e-09         27.460253         27.774704
99  0.697162     100 -4.860350e-01         27.768153         28.142431

Table of Contents#

API Reference

Indices and tables#