thuijskens / scikit-hyperband Goto Github PK

If I want to set a total budget for experiments just like the original paper, what should I do?

self.fit_params is deprecated in scikit-learn 0.21.0 ( https://github.com/scikit-learn/scikit-learn/commit/#13519 ). scikit-optimize also has the similar issue and you can read more in this page.

When I run this,

model = lgb.LGBMClassifier()
param_dist = {
    'max_depth': [3, 5,10]
}
search = HyperbandSearchCV(model, param_dist, 
                           resource_param='n_estimators',
                           scoring='roc_auc')

I got the error like this.

TypeError: __init__() got an unexpected keyword argument 'fit_params'

Below are the warnings and errors I got when doing python setup.py install. Making this pip install-able would be very helpful.

SetuptoolsDeprecationWarning: setup.py install is deprecated. Use build and pip and other standards-based tools.
  warnings.warn(

EasyInstallDeprecationWarning: easy_install command is deprecated. Use build and pip and other standards-based tools.
  warnings.warn(

error: can't create or remove files in install directory

The following error occurred while trying to add or remove files in the
installation directory:

    [Errno 13] Permission denied: 'C:\\Program Files\\WindowsApps\\PythonSoftwareFoundation.Python.3.9_3.9.3568.0_x64__qbz5n2kfra8p0\\Lib\\site-packages\\test-easy-install-10696.write-test'

Hello,

I have always an error
ValueError: Invalid parameter n_estimators for estimator Pipeline(memory=None,
steps=[('umap', UMAP(a=None, angular_rp_forest=False, b=None, init='spectral',
learning_rate=1.0, local_connectivity=1.0, metric='euclidean',
metric_kwds=None, min_dist=0.1, n_components=2, n_epochs=None,
n_neighbors=15, negative_sample_rate=5, random_state=456,
repulsion_strength=1.0, s...ax_iter=1000,
multi_class='ovr', penalty='l2', random_state=None, tol=0.0001,
verbose=0))]). Check the list of available parameters with estimator.get_params().keys().

my code
from sklearn.model_selection import RandomizedSearchCV
umap2 = umap.UMAP(random_state=456)
svc=LinearSVC(dual=False)
pipeline = Pipeline([("umap", umap2),
("svc", svc)])

params_grid_pipeline = {"umap__n_neighbors": [5,20],
"umap__n_components": [15, 25, 50],
"svc__C": [10**k for k in range(-3, 4)],
'svc__random_state':[43]}

gs=HyperbandSearchCV(estimator=pipeline, param_distributions=params_grid_pipeline, scoring='recall', cv=10, verbose=1,n_jobs = 5)

gs = gs.fit(X_train.values,y_train.values.ravel())

My params are good

Hi! Can I use HyperbandSearchCV for tuning sklearn.neural_network?
I don't know how to set the parameter 'resource_param' in this situation.
Thx!

Receives similar error to scikit-optimize/scikit-optimize#978

TypeError: init() got an unexpected keyword argument 'iid'

great code thanks and this is great too
https://www.youtube.com/watch?v=JsArBz46_3s&index=20&list=PLGVZCDnMOq0ovNxfxOqYcBcQOIny9Zvb-&t=22s
Why giving your algorithm ALL THE FEATURES does not always work - Thomas Huijskens

but can you help find code in python for FEATURES selection

for example this is in C
https://github.com/Craigacp/FEAST

and this
https://github.com/scikit-learn-contrib/stability-selection
dose not use mutual info

Hi, Thank you for this excellent work. I was wondering if you could add the ability to send eval_set as a fit_params in search.py? This will allow the fit method to do early stopping (which currently isn't possible using sklearn gridsearchcv). I updated the search.py code in line 578-580 marked by 3 hash marks below:

                    for train, test in cv.split(X, y, groups):
                        X_test = X.iloc[test.tolist()]  ###
                        y_test = y[test.tolist()] ###
                        fit_params = {"eval_set": (X_test, y_test)} ###
                        jobs.append(delayed(_fit_and_score)(
                            clone(base_estimator), X, y, scorers,
                            train, test, self.verbose, parameters,
                            fit_params=fit_params,
                            return_train_score=self.return_train_score,
                            return_n_test_samples=True,
                            return_times=True, return_parameters=True,
                            error_score=self.error_score
                        ))

I know the above is a hack which is why maybe you could figure out a better way to do something like this in your package?

WIth the above edit, I can tune catboost using the CV folds for early stopping. As opposed to gridsearchcv where you have to pass in a fixed test set.

Can we put this on pypi so that others can easily download it? It's a little janky to download and install from repo.

top_configurations = [x for , x in sorted(zip(results['rank_test%s' % refit_metric],
results['params']),
key=lambda x: x[0])]

In the above code, when we want to get the top_configurations of one bracket, we, in fact, take all brackets into consideration. is it right?