def test_scenario8(self): """ Scenario: Successfully building a new dataset from an existing one and analyzing it Given I create a BigML dataset from "<data>" and store logs in "<output_dir>" And I check that the source has been created And I check that the dataset has been created And I create a new BigML dataset using the specs in JSON file "<new_fields>" and a model with "<model_fields>" And I check that the new dataset has been created And I check that the model has been created And I create BigML nodes analysis from <min_nodes> to <max_nodes> by <nodes_step> with <kfold>-cross-validation improving "<metric>" And I check that the <kfold>-datasets have been created And I check that the <kfold>-models have been created And I check that all the <kfold>-fold cross-validations have been created Then the best node threshold is "<node_threshold>", with "<metric>" of <metric_value> Examples: |data |output_dir |new_fields | field | model_fields| min_nodes | max_nodes | nodes_step | kfold | metric | node_threshold | metric_value | |../data/iris.csv | ./scenario_a_10 |../data/new_fields.json| outlier? |petal length,outlier?,species| 3 | 14 | 2 |2 | precision | 9 | 94.71% | """ print self.test_scenario8.__doc__ examples = [ ['data/iris.csv', 'scenario_a_10', 'data/new_fields2.json', u'outlier?', u'outlier?,species', '3', '14', '2', '2', 'precision', '5', '98.21%']] for example in examples: print "\nTesting with:\n", example dataset_adv.i_create_dataset(self, data=example[0], output_dir=example[1]) test_pred.i_check_create_source(self) test_pred.i_check_create_dataset(self, suffix=None) dataset_adv.i_create_dataset_new_fields(self, json_file=example[2], model_fields=example[4]) test_pred.i_check_create_new_dataset(self) test_pred.i_check_create_model(self) test_pred.i_create_nodes_analysis(self, min_nodes=example[5], max_nodes=example[6], nodes_step=example[7], k_fold=example[8], metric=example[9]) test_pred.i_check_create_kfold_datasets(self, example[8]) test_pred.i_check_create_kfold_models(self, example[8]) test_pred.i_check_create_all_kfold_cross_validations(self, example[8]) test_pred.i_check_node_threshold(self, example[10], example[9], example[11])
def test_scenario5(self): """ Scenario: Successfully building nodes threshold analysis from dataset file: Given I create BigML dataset uploading train "<data>" file in "<output>" And I check that the source has been created And I check that the dataset has been created And I create BigML nodes analysis from dataset file from <min_nodes> to <max_nodes> by <nodes_step> with <kfold>-cross-validation improving "<metric>" And I check that the <kfold>-datasets have been created And I check that the <kfold>-models have been created And I check that all the <kfold>-fold cross-validations have been created Then the best node threshold is "<node_threshold>", with "<metric>" of <metric_value> Examples: | data | output | min_nodes | max_nodes | nodes_step | kfold | metric | node_threshold | metric_value | ../data/iris.csv | ./scenario_a_4/evaluation | 3 | 14 | 2 |2 | precision | 9 | 94.71% """ print self.test_scenario5.__doc__ examples = [ ['data/iris.csv', 'scenario_a_4/evaluation', '3', '14', '2', '2', 'precision', '9', '94.71%']] for example in examples: print "\nTesting with:\n", example test_pred.i_create_dataset(self, data=example[0], output=example[1]) test_pred.i_check_create_source(self) test_pred.i_check_create_dataset(self) test_pred.i_create_nodes_analysis_from_dataset_file( self, min_nodes=example[2], max_nodes=example[3], nodes_step=example[4], k_fold=example[5], metric=example[6]) test_pred.i_check_create_kfold_datasets(self, example[5]) test_pred.i_check_create_kfold_models(self, example[5]) test_pred.i_check_create_all_kfold_cross_validations(self, example[5]) test_pred.i_check_node_threshold(self, example[7], example[6], example[8])