tuun

KO | EN

gitlite — search

#javascript #react #llm #golang #docker #macos #windows #cli #kubernetes #awesome-list #security #mcp #openai #open-source

(★ 51)

Hyperparameter tuning via uncertainty modeling

File Explorer

Download Latest Version (.zip)

- - release_pypi.yaml
  - test.yaml
- - benchmark.rst
  - index.rst
- - branin40.svg
  - casl_logo.svg
  - cmu_logo.png
  - github.svg
  - hartmann6.svg
  - petuum_logo.svg
  - stanford_logo.png
  - tuun_logo.png
  - tuun_logo.svg
  - tuun_logo_only.png
  - tuun_logo_space.png
- at2.rst
- conf.py
- configure.rst
- index.rst
- installation.rst
- make.bat
- Makefile
- nni.rst
- quick_start.rst
- search_space.rst
- - 00_stan.py
  - 01_stan_config.py
  - branin.py
- - 00_stan.py
- - 00_stan.py
  - hartmann.py
- - branin.py
  - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - branin.py
  - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - run_nni.sh
  - search_space.json
  - search_space1.json
- - augment.py
  - main.py
  - nni_config.yml
  - README.md
  - requirements.txt
  - resnet.py
  - search_space.json
- - config.yml
  - hartmann.py
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - run_nni.sh
  - search_space.json
  - trial.py
- - config.yml
  - data_setup.py
  - lib.py
  - run_nni.sh
  - search_space.json
  - search_space2.json
  - trial_nni.py
- - plot_benchmark.py
  - plot_benchmark_old.py
  - plot_cifar10.py
  - plot_xgboost.py
- - minimize_01.py
  - minimize_02.py
  - minimize_03.py
  - minimize_04.py
  - minimize_05.py
  - minimize_06.py
  - minimize_07.py
  - minimize_08.py
  - minimize_09.py
  - minimize_10.py
  - minimize_11.py
- - 00_minimal_example.py
  - 01_minimal_example_2d.py
  - 02_minimal_bo.py
  - 03a_probo_backend_suggest.py
  - 03b_probo_backend_suggest.py
  - 04a_probo_backend_minimize.py
  - 04b_probo_backend_minimize.py
  - 05a_dragonfly_backend_suggest.py
  - 05b_dragonfly_backend_suggest.py
  - 06_dragonfly_backend_minimize.py
  - 07_probo_backend_suggest_product.py
  - 08a_probo_backend_minimize_product.py
  - 08b_probo_backend_minimize_product.py
  - 08c_probo_backend_minimize_product.py
  - 08d_probo_backend_minimize_product.py
  - 09_dragonfly_backend_suggest_product.py
  - 10_dragonfly_backend_minimize_product.py
  - 11a_tuun_suggest.py
  - 11b_tuun_suggest.py
  - 12a_tuun_minimize.py
  - 12b_tuun_minimize.py
  - 12c_tuun_minimize.py
  - 12d_tuun_minimize.py
  - 12e_tuun_minimize.py
  - 12f_tuun_minimize.py
  - 13a_tuun_minimize.py
  - 14a_tuun_minimize.py
  - 14b_tuun_minimize.py
  - 14c_tuun_minimize.py
  - 14d_tuun_minimize.py
  - 15a_tuun_minimize.py
  - 15b_tuun_minimize.py
  - 16a_tuun_minimize.py
  - 16b_tuun_minimize.py
  - 16c_tuun_minimize.py
  - 16d_tuun_minimize.py
- - data_loader.py
  - minimize_config_list.py
  - minimize_example_transfertune.py
  - minimize_example_vanillatune.py
  - parse_online_task.py
- requirements.txt
- requirements_all.txt
- requirements_dev.txt
- requirements_dragonfly.txt
- requirements_gpytorch.txt
- requirements_nni.txt
- requirements_stan.txt
- add_pwd_to_pythonpath.sh
- clean.sh
- copy_all_tuun_files.sh
- format_tuun_black.sh
- print_files_containing_string.sh
- run_all_tests.sh
- - image_classification.tgz
  - info.json
- - info.json
  - question_answering.tgz
- - test_dragonflybackend.py
  - test_probobackend.py
- - test_integral.py
  - test_list.py
  - test_real.py
- - __init__.py
  - test_logging.py
- __init__.py
- test_tuun.py
- - __init__.py
  - acqdef.py
  - acqfun.py
  - acqopt.py
  - acqopt_spo.py
- - __init__.py
  - at2.py
  - core.py
  - dragonfly.py
  - probo.py
  - regression_model.py
- - __init__.py
  - bo.py
- - __init__.py
  - integral.py
  - list.py
  - product.py
  - real.py
- - __init__.py
  - gp_gpytorch.py
  - gp_simple.py
  - gp_stan.py
  - penn_sklearn.py
  - README.md
- - __init__.py
- - __init__.py
  - data_transform.py
  - database.py
  - log.py
  - misc_util.py
- __init__.py
- main.py
- nni_tuner.py
.gitignore
hubconf.py
LICENSE
README.md
setup.py

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

&emsp;&emsp;
 <a href="https://petuum.github.io/tuun">Documentation</a> and
 <a href="https://github.com/petuum/tuun/tree/master/examples">Examples</a>

**Tuun** is a toolkit for efficient hyperparameter tuning via uncertainty
modeling, with a focus on flexible model choice, scalability, and use in
distributed settings.

## Installation

Tuun requires Python 3.6+. To install all dependencies for development, clone and `cd`
into this repo, and run:
```
$ pip install -r requirements/requirements_dev.txt
```

For the full functionality of Tuun, a [Stan](https://mc-stan.org/) model must also be
compiled (this takes roughly 1 minute) by running:
```
$ python tuun/probo/models/stan/compile_models.py -m gp_distmat_fixedsig
```

## Quick Start

**Note:** Tuun is still in the early stages and the following API and functionality may
undergo changes.

[Here](examples/quick_start/minimize_01.py) is a minimal working example, which uses
Tuun to optimize a function over a one-dimensional Euclidean search space with bounds
[-5, 5].

```python
from tuun.main import Tuun

# instantiate Tuun
tu = Tuun()

# set search space
search_space = ('real', [-5, 5])
tu.set_config_from_list(search_space)

# define function to optimize
f = lambda x: x[0] ** 4 - x[0] ** 2 + 0.1 * x[0]

# minimize function over search space
result = tu.minimize_function(f, 20)
```
This should find a minima at roughly: 𝑥\*=[−0.73], where 𝑓(𝑥\*)=−0.32.  See [this docs
page](https://petuum.github.io/tuun/search_space.html) for more details on defining
different search spaces for Tuun.

Tuun also allows for fine-grained configuration of individual components and search
spaces, as shown in the [example](examples/quick_start/minimize_11.py) below.

```python
from tuun.main import Tuun

config = {
    # configure tuning backend
    'backend': 'probo',

# configure model
    'model_config': {'name': 'simplegp', 'ls': 3.0, 'alpha': 1.5, 'sigma': 1e-5},

# configure acquisition function
    'acqfunction_config': {'name': 'default', 'acq_str': 'ei'},

# configure acquisition optimizer
    'acqoptimizer_config': {'name': 'default', 'max_iter': 200},

# configure domain
    'domain_config': ('real', [-5, 5]),
}
tu = Tuun(config)

f = lambda x: x[0] ** 4 - x[0] ** 2 + 0.1 * x[0]
result = tu.minimize_function(f, 20)
```
This should also find a minima at roughly: 𝑥\*=[−0.73], where 𝑓(𝑥\*)=−0.32. See [this
docs page](https://petuum.github.io/tuun/configure.html) for more details on possible
configurations.

[See here](examples/quick_start/) for a larger set of quick start examples for Tuun.

## Use in NNI

Tuun can be used as a custom tuner in [NNI (neural network
intelligence)](https://github.com/microsoft/nni), which allows for visualization and
experiment management. See [this docs page](https://petuum.github.io/tuun/nni.html) for
more details, and [this directory](examples/nni_simple_2d) for a minimal example.

## Examples

See a few examples of Tuun [here](examples/), including the setup for various benchmark
functions, different types of search spaces, and use within NNI.

The plots below show a couple examples of Tuun, along with other tuning algorithms, on
benchmark functions.

 <img src="docs/images/hartmann6.svg" alt="Hartmann 6 Dimensions" width="48%">
 &nbsp; &nbsp;
 <img src="docs/images/branin40.svg" alt="Branin 40 Dimensions" width="48%">

## Affiliations

Tuun is part of the [CASL project](http://casl-project.ai/).
<img src="docs/images/casl_logo.svg" width=280>

Companies and universities using and developing Tuun.

 <img src="docs/images/cmu_logo.png" width="15%">
 &nbsp; &nbsp; &nbsp; &nbsp;
 <img src="docs/images/petuum_logo.svg" width="24%">
 &nbsp; &nbsp;
 <img src="docs/images/stanford_logo.png" width="20%">

tuun

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

tuun

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

Download README