Welcome to NePS, a powerful and flexible Python library for hyperparameter optimization (HPO) and neural architecture search (NAS) with its primary goal: make HPO and NAS usable for deep learners in practice.
NePS houses recently published and also well-established algorithms that can all be run massively parallel on distributed setups, with tools to analyze runs, restart runs, etc., all tailored to the needs of deep learning experts.
Take a look at our documentation for all the details on how to use NePS!
In addition to the features offered by traditional HPO and NAS libraries, NePS, e.g., stands out with:
- Hyperparameter Optimization (HPO) with Prior Knowledge and Cheap Proxies:
NePS excels in efficiently tuning hyperparameters using algorithms that enable users to make use of their prior knowledge within the search space. This is leveraged by the insights presented in: - Neural Architecture Search (NAS) with General Search Spaces:
NePS is equipped to handle context-free grammar search spaces, providing advanced capabilities for designing and optimizing architectures. this is leveraged by the insights presented in: - Easy Parallelization and Design Tailored to DL:
NePS simplifies the process of parallelizing optimization tasks both on individual computers and in distributed computing environments. As NePS is made for deep learners, all technical choices are made with DL in mind and common DL tools such as Tensorboard are embraced.
To install the latest release from PyPI run
pip install neural-pipeline-search
To get the latest version from Github run
pip install git+https://github.com/automl/neps.git
Using neps
always follows the same pattern:
- Define a
run_pipeline
function capable of evaluating different architectural and/or hyperparameter configurations for your problem. - Define a search space named
pipeline_space
of those Parameters e.g. via a dictionary - Call
neps.run
to optimizerun_pipeline
overpipeline_space
In code, the usage pattern can look like this:
import neps
import logging
# 1. Define a function that accepts hyperparameters and computes the validation error
def run_pipeline(
hyperparameter_a: float, hyperparameter_b: int, architecture_parameter: str
) -> dict:
# Create your model
model = MyModel(architecture_parameter)
# Train and evaluate the model with your training pipeline
validation_error = train_and_eval(
model, hyperparameter_a, hyperparameter_b
)
return validation_error
# 2. Define a search space of parameters; use the same parameter names as in run_pipeline
pipeline_space = dict(
hyperparameter_a=neps.FloatParameter(
lower=0.001, upper=0.1, log=True # The search space is sampled in log space
),
hyperparameter_b=neps.IntegerParameter(lower=1, upper=42),
architecture_parameter=neps.CategoricalParameter(["option_a", "option_b"]),
)
# 3. Run the NePS optimization
logging.basicConfig(level=logging.INFO)
neps.run(
run_pipeline=run_pipeline,
pipeline_space=pipeline_space,
root_directory="path/to/save/results", # Replace with the actual path.
max_evaluations_total=100,
)
Discover how NePS works through these examples:
-
Hyperparameter Optimization: Learn the essentials of hyperparameter optimization with NePS.
-
Multi-Fidelity Optimization: Understand how to leverage multi-fidelity optimization for efficient model tuning.
-
Utilizing Expert Priors for Hyperparameters: Learn how to incorporate expert priors for more efficient hyperparameter selection.
-
Architecture Search: Dive into (hierarchical) architecture search in NePS.
-
Additional NePS Examples: Explore more examples, including various use cases and advanced configurations in NePS.
Please see the documentation for contributors.
For pointers on citing the NePS package and papers refer to our documentation on citations.