def rgf_cv(X, y, n_folds, cv_params):
L2 = cv_params['L2']
sL2 = cv_params['sL2']
reg_depth = cv_params['reg_depth']
algorithm = cv_params['algorithm']
loss = cv_params['loss']
max_leaf_forest = cv_params['max_leaf_forest']
test_interval = cv_params['test_interval']
simdir = cv_params['simdir']
n, p = X.shape
# create cv number of files for cross validation
kf = cross_validation.KFold(n, n_folds=n_folds,
shuffle=True,
random_state=1234)
ll = []
ncv = 0
for train_index, test_index in kf:
simdir_cv = simdir + '/' + str(ncv)
X_train, X_test = X[train_index, :], X[test_index, :]
y_train, y_test = y[train_index], y[test_index]
mkdir_p(simdir+'/data')
np.savetxt(simdir + '/data/y_test-%d' % ncv, y_test)
rg = RGFC(simdir=simdir_cv, algorithm=algorithm, loss=loss,
reg_L2=L2, reg_sL2=sL2, reg_depth=reg_depth,
max_leaf_forest=max_leaf_forest, test_interval=test_interval,
n_labels=n_labels, n_jobs=-1)
rg.fit(X_train, y_train)
proba = rg.predict_proba(X_test)
ll.append(ru.calc_ll_from_proba(proba, y_test))
ncv += 1
ll = np.array(ll)
return ll.mean(), ll.std()
