0 votes
1 view
in Machine Learning by (17.4k points)

I am running into the problem that the hyperparameters of my svm.SVC() are too wide such that the GridSearchCV() never gets completed! One idea is to use RandomizedSearchCV() instead. But again, my dataset is relative big such that 500 iterations take about 1 hour!

My question is, what is a good set-up (in terms of the range of values for each hyperparameter) in GridSearchCV ( or RandomizedSearchCV ) in order to stop wasting resources?

In other words, how to decide whether or not e.g. C values above 100 make sense and/or step of 1 is neither big not small? Any help is very much appreciated. This is the set-up am currently using:

parameters = {

    'C':            np.arange( 1, 100+1, 1 ).tolist(),

    'kernel':       ['linear', 'rbf'],                   # precomputed,'poly', 'sigmoid'

    'degree':       np.arange( 0, 100+0, 1 ).tolist(),

    'gamma':        np.arange( 0.0, 10.0+0.0, 0.1 ).tolist(),

    'coef0': np.arange( 0.0, 10.0+0.0, 0.1 ).tolist(),

    'shrinking':    [True], 'probability':  [False],

    'tol': np.arange( 0.001, 0.01+0.001, 0.001 ).tolist(),

    'cache_size':   [2000],

    'class_weight': [None],

    'verbose':      [False],

    'max_iter':     [-1],

    'random_state': [None],

    }

model = grid_search.RandomizedSearchCV( n_iter              = 500,

estimator = svm.SVC(),

                                        param_distributions = parameters, n_jobs = 4, iid = True, refit = True, cv = 5, verbose = 1, pre_dispatch  = '2*n_jobs')         

# scoring = 'accuracy'

model.fit( train_X, train_Y )

print( model.best_estimator_ )

print( model.best_score_ )

print( model.best_params_ )

1 Answer

0 votes
by (33.2k points)

The choice of kernel depends on your data, the number of samples and dimensions. For the ranges to be comparable, you need to normalize your data, often StandardScaler, which does zero mean and unit variance, is a good idea. If your data is non-negative, then try MinMaxScaler.

For kernel="gamma"

{'C': np.logspace(-3, 2, 6), 'gamma': np.logspace(-3, 2, 6)}

One of the main advantages of a randomized search is that you can actually search continuous parameters using continuous distributions.

Hope this answer helps you! For more details, check the Machine Learning Algorithms.

Also, undergoing a Machine Learning Certification would pave a way for better knowledge.

Welcome to Intellipaat Community. Get your technical queries answered by top developers !


Categories

...