Grid search hyper-parameter optimization using a validation set (not cross validation)
Project description
A Python machine learning package for grid search hyper-parameter optimization using a validation set (defaults to cross validation when no validation set is available). This package works for Python 2.7+ and Python 3+, for any model (classification and regression), and runs in parallel on all threads on your CPU automatically.
scikit-learn provides a package for grid-search hyper-parameter optimization **using cross-validation** on the training dataset. Unfortunately, cross-validation is impractically slow for large datasets and fails for small datasets due to the lack of data in each class needed to properly train each fold. Instead, we use a constant validation set to optimize hyper-parameters – the hypopt package makes this fast (distributed on all CPU threads) and easy (one line of code).
hypopt.model_selection.fit_model_with_grid_search supports grid search hyper-parameter optimization when you already have a validation set, eliminating the extra hours of training time required when using cross-validation. However, when no validation set is given, it defaults to using cross-validation on the training set. This allows you to alows use hypopt anytime you need to do hyper-parameter optimization with grid-search, regardless of whether you use a validation set or cross-validation.
Installation
Python 2.7, 3.4, 3.5, and 3.6 are supported.
Stable release:
$ pip install hypopt
Developer (unstable) release:
$ pip install git+https://github.com/cgnorthcutt/hypopt.git
To install the codebase (enabling you to make modifications):
$ conda update pip # if you use conda
$ git clone https://github.com/cgnorthcutt/hypopt.git
$ cd hypopt
$ pip install -e .
Examples
Basic usage
# Assuming you already have train, test, val sets and a model.
from hypopt import GridSearch
param_grid = [
{'C': [1, 10, 100], 'kernel': ['linear']},
{'C': [1, 10, 100], 'gamma': [0.001, 0.0001], 'kernel': ['rbf']},
]
# Grid-search all parameter combinations using a validation set.
gs = GridSearch(model = SVR())
gs.fit(X_train, y_train, param_grid, X_val, y_val)
print('Test Score for Optimized Parameters:', gs.score(X_test, y_test))
Minimal working examples
Other Examples including a working example with MNIST
Use hypopt with any model (PyTorch, Tensorflow, caffe2, scikit-learn, etc.)
All of the features of the hypopt package work with any model. Yes, any model. Feel free to use PyTorch, Tensorflow, caffe2, scikit-learn, mxnet, etc. If you use a scikit-learn model, all hypopt methods will work out-of-the-box. It’s also easy to use your favorite model from a non-scikit-learn package, just wrap your model into a Python class that inherets the sklearn.base.BaseEstimator. Here’s an example for a generic classifier:
from sklearn.base import BaseEstimator
class YourModel(BaseEstimator): # Inherits sklearn base classifier
def __init__(self, ):
pass
def fit(self, X, y, sample_weight = None):
pass
def predict(self, X):
pass
def score(self, X, y, sample_weight = None):
pass
# Inherting BaseEstimator gives you these for free!
# So if you inherit, there's no need to implement these.
def get_params(self, deep = True):
pass
def set_params(self, **params):
pass
PyTorch MNIST CNN Example
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for hypopt-1.0.3-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | b7f7f48e2c4ee7c7c816b80de705c3b14ae832be882e3425e1265e31a2b17cef |
|
MD5 | 1e4de4fee6efc6d34bde9274529c53d1 |
|
BLAKE2b-256 | 5a291ae04b420825f0b804ff7112dbd109e209a6a47c67e6b83f8667592ee343 |