Skip to content

Instantly share code, notes, and snippets.

@mnarayan

mnarayan/skggm_optimal_selection_example.py

Created October 8, 2017 17:43
Show Gist options
  • Save mnarayan/1c4287e3b8f8105c218c2150a8e5e7b4 to your computer and use it in GitHub Desktop.
Save mnarayan/1c4287e3b8f8105c218c2150a8e5e7b4 to your computer and use it in GitHub Desktop.
Download ZIP
Example of using scikit-optimize with skggm
Raw
skggm_optimal_selection_example.py
import sys
import numpy as np
from skopt.space import Real, Categorical, Integer
from skopt import BayesSearchCV
from sklearn.grid_search import GridSearchCV
from sklearn.datasets import make_sparse_spd_matrix
from sklearn.model_selection import train_test_split
from sklearn.metrics import make_scorer
sys.path.append('..')
from inverse_covariance import (
QuicGraphLasso,
QuicGraphLassoCV,
)
def make_data(n_samples, n_features):
prng = np.random.RandomState(1)
prec = make_sparse_spd_matrix(
n_features,
alpha=.98,
smallest_coef=.4,
largest_coef=.7,
random_state=prng
)
cov = np.linalg.inv(prec)
d = np.sqrt(np.diag(cov))
cov /= d
cov /= d[:, np.newaxis]
prec *= d
prec *= d[:, np.newaxis]
X = prng.multivariate_normal(np.zeros(n_features), cov, size=n_samples)
X -= X.mean(axis=0)
X /= X.std(axis=0)
return X, cov, prec
def quic_graph_lasso_gridsearch(X, num_folds, metric):
'''Run QuicGraphLasso with mode='default' and use standard scikit
GridSearchCV to find the best lambda.
Primarily demonstrates compatibility with existing scikit tooling.
'''
print('QuicGraphLasso + GridSearchCV with:')
print(' metric: {}'.format(metric))
search_grid = {
'lam': np.logspace(
np.log10(0.01), np.log10(1.0), num=100, endpoint=True
)
}
model = GridSearchCV(QuicGraphLasso(init_method='corrcoef',score_metric=metric),
search_grid,
cv=num_folds,
refit=True)
model.fit(X)
bmodel = model.best_estimator_
print(' len(cv_lams): {}'.format(len(search_grid['lam'])))
print(' cv-lam: {}'.format(model.best_params_['lam']))
print(' lam_scale_: {}'.format(bmodel.lam_scale_))
print(' lam_: {}'.format(bmodel.lam_))
return bmodel.covariance_, bmodel.precision_, bmodel.lam_
def quic_graph_lasso_skopt(X, num_folds, metric):
'''Run QuicGraphLasso with mode='default' and use standard scikit
GridSearchCV to find the best lambda.
Primarily demonstrates compatibility with existing scikit tooling.
'''
print('QuicGraphLasso + skopt.BayesSearchCV with:')
print(' metric: {}'.format(metric))
model = BayesSearchCV(QuicGraphLasso(init_method='corrcoef', score_metric=metric),
cv=num_folds,
n_iter=16,
refit=True
) # Using n_iter parameter seems to mess up parameter search space
model.add_spaces('space_1', {
'lam': Real(1e-02,1e+1, prior='log-uniform')
}
)
model.fit(X)
score = model.score(X)
print(' Final score: {}'.format(score))
bmodel = model.best_estimator_
print(' cv-lam: {}'.format(model.best_params_['lam']))
print(' lam_scale_: {}'.format(bmodel.lam_scale_))
print(' lam_: {}'.format(bmodel.lam_))
# X_train, X_test, = train_test_split(X,train_size=0.5, random_state=0)
# for i in range(4):
# model.step(X,None,'space_1')
# # save the model or use custom stopping criterion here
# # model is updated after every step
# # ...
# score = model.score(X)
# print(' step: {},score: {}'.format(i, score))
# bmodel = model.best_estimator_
# print(' cv-lam: {}'.format(model.best_params_['lam']))
# print(' lam_scale_: {}'.format(bmodel.lam_scale_))
# print(' lam_: {}'.format(bmodel.lam_))
return bmodel.covariance_, bmodel.precision_, bmodel.lam_
n_samples = 150;
n_features = 10
X, true_cov, true_prec = make_data(n_samples,n_features)
quic_graph_lasso_gridsearch(X,2,'kl')
quic_graph_lasso_skopt(X,2,'kl')
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment