def test_5000_classes(self): weights = ([0.0002] * 4750) + ([0.0001] * 250) X, Y = sklearn.datasets.make_classification(n_samples=10000, n_features=20, n_classes=5000, n_clusters_per_class=1, n_informative=15, n_redundant=5, n_repeated=0, weights=weights, flip_y=0, class_sep=1.0, hypercube=True, shift=None, scale=1.0, shuffle=True, random_state=1) self.assertEqual(250, np.sum(np.bincount(Y) == 1)) D = Dummy() D.info = {'metric': 'r2_metric', 'task': MULTICLASS_CLASSIFICATION, 'is_sparse': False, 'target_num': 1} D.data = {'X_train': X, 'Y_train': Y, 'X_valid': X, 'X_test': X} D.feat_type = ['numerical'] * 5000 configuration_space = get_configuration_space(D.info, include_estimators=['extra_trees'], include_preprocessors=['no_preprocessing']) configuration = configuration_space.sample_configuration() D_ = copy.deepcopy(D) evaluator = HoldoutEvaluator(D_, configuration) evaluator.fit()
def main(dataset_info, mode, seed, params, mode_args=None): """This command line interface has three different operation modes: * CV: useful for the Tweakathon * 1/3 test split: useful to evaluate a configuration * cv on 2/3 train split: useful to optimize hyperparameters in a training mode before testing a configuration on the 1/3 test split. It must by no means be used for the Auto part of the competition! """ if mode != "test": num_run = get_new_run_num() for key in params: try: params[key] = int(params[key]) except: try: params[key] = float(params[key]) except: pass if seed is not None: seed = int(float(seed)) else: seed = 1 output_dir = os.getcwd() D = store_and_or_load_data(dataset_info=dataset_info, outputdir=output_dir) cs = get_configuration_space(D.info) configuration = configuration_space.Configuration(cs, params) metric = D.info['metric'] global evaluator # Train/test split if mode == 'holdout': evaluator = HoldoutEvaluator(D, configuration, with_predictions=True, all_scoring_functions=True, output_y_test=True, seed=seed, num_run=num_run) evaluator.fit() signal.signal(15, empty_signal_handler) evaluator.finish_up() model_directory = os.path.join(os.getcwd(), "models_%d" % seed) if os.path.exists(model_directory): model_filename = os.path.join(model_directory, "%s.model" % num_run) with open(model_filename, "w") as fh: pickle.dump(evaluator.model, fh, -1) elif mode == 'test': evaluator = TestEvaluator(D, configuration, all_scoring_functions=True, seed=seed) evaluator.fit() scores = evaluator.predict() duration = time.time() - evaluator.starttime score = scores[metric] additional_run_info = ";".join( ["%s: %s" % (m_, value) for m_, value in scores.items()]) additional_run_info += ";" + "duration: " + str(duration) print "Result for ParamILS: %s, %f, 1, %f, %d, %s" % ( "SAT", abs(duration), score, evaluator.seed, additional_run_info) # CV on the whole training set elif mode == 'cv': evaluator = CVEvaluator(D, configuration, with_predictions=True, all_scoring_functions=True, output_y_test=True, cv_folds=mode_args['folds'], seed=seed, num_run=num_run) evaluator.fit() signal.signal(15, empty_signal_handler) evaluator.finish_up() elif mode == 'partial_cv': evaluator = CVEvaluator(D, configuration, all_scoring_functions=True, cv_folds=mode_args['folds'], seed=seed, num_run=num_run) evaluator.partial_fit(mode_args['fold']) scores = evaluator.predict() duration = time.time() - evaluator.starttime score = scores[metric] additional_run_info = ";".join( ["%s: %s" % (m_, value) for m_, value in scores.items()]) additional_run_info += ";" + "duration: " + str(duration) print "Result for ParamILS: %s, %f, 1, %f, %d, %s" % ( "SAT", abs(duration), score, evaluator.seed, additional_run_info) elif mode == 'nested-cv': evaluator = NestedCVEvaluator(D, configuration, with_predictions=True, inner_cv_folds=mode_args['inner_folds'], outer_cv_folds=mode_args['outer_folds'], all_scoring_functions=True, output_y_test=True, seed=seed, num_run=num_run) evaluator.fit() signal.signal(15, empty_signal_handler) evaluator.finish_up() else: raise ValueError("Must choose a legal mode.")