Example #1
0
    def test_scenario6(self):
        """
            Scenario 6: Successfully creating a fusion:
                Given I create a data source uploading a "<data>" file
                And I wait until the source is ready less than <time_1> secs
                And I create a dataset
                And I wait until the dataset is ready less than <time_2> secs
                And I create a model with "<params>"
                And I wait until the model is ready less than <time_3> secs
                And I create a logistic regression with "<params>"
                And I wait until the logistic regression is ready less than <time_3> secs
                And I create a logistic regression with "<params>"
                And I wait until the logistic regression is ready less than <time_3> secs
                And I retrieve a list of remote logistic regression tagged with "<tag>"
                And I create a fusion from a list of models and weights
                And I wait until the fusion is ready less than <time_4> secs
                When I create a prediction for "<data_input>"
                Then the prediction for "<objective>" is "<prediction>"
                And the fusion probability for the prediction is "<probability>"
                And I create a local fusion prediction for "<data_input>"
                Then the local fusion prediction is "<prediction>"
                And the local fusion probability for the prediction is "<probability>"

                Examples:
                | data                | time_1  | time_2 | time_3 | time_4 | data_input | objective | prediction
                | ../data/iris.csv | 10      | 10     | 20     | 20 | {"petal length": 1, "petal width": 1} | "000004" | "Iris-setosa"
        """
        print self.test_scenario6.__doc__
        examples = [
            ['data/iris.csv', '10', '10', '20', '20',
             '{"tags":["my_fusion_6_tag"], "missing_numerics": true}',
             'my_fusion_6_tag',
             '{"petal width": 1.75, "petal length": 2.45}',
             "000004",
             "Iris-setosa",
             '0.4727',
             '{"tags":["my_fusion_6_tag"], "missing_numerics": false, "balance_fields": false }', '[1, 2]']]
        for example in examples:
            print "\nTesting with:\n", example
            source_create.i_upload_a_file(self, example[0])
            source_create.the_source_is_finished(self, example[1])
            dataset_create.i_create_a_dataset(self)
            dataset_create.the_dataset_is_finished_in_less_than(self, example[2])
            model_create.i_create_a_logistic_model_with_objective_and_parms(self, example[8], example[5])
            model_create.the_logistic_model_is_finished_in_less_than(self, example[3])
            model_create.i_create_a_logistic_model_with_objective_and_parms(self, example[8], example[11])
            model_create.the_logistic_model_is_finished_in_less_than(self, example[3])
            compare_pred.i_retrieve_a_list_of_remote_logistic_regressions(self, example[6])
            model_create.i_create_a_fusion_with_weights(self, example[12])
            model_create.the_fusion_is_finished_in_less_than(self, example[3])
            compare_pred.i_create_a_local_fusion(self)
            prediction_create.i_create_a_fusion_prediction(self, example[7])
            prediction_create.the_prediction_is(self, example[8], example[9])
            prediction_create.the_fusion_probability_is(self, example[10])
            compare_pred.i_create_a_local_prediction(self, example[7])
            compare_pred.the_local_prediction_is(self, example[9])
            compare_pred.the_local_probability_is(self, example[10])
    def test_scenario6(self):
        """
            Scenario 6: Successfully creating a fusion:
                Given I create a data source uploading a "<data>" file
                And I wait until the source is ready less than <time_1> secs
                And I create a dataset
                And I wait until the dataset is ready less than <time_2> secs
                And I create a model with "<params>"
                And I wait until the model is ready less than <time_3> secs
                And I create a logistic regression with "<params>"
                And I wait until the logistic regression is ready less than <time_3> secs
                And I create a logistic regression with "<params>"
                And I wait until the logistic regression is ready less than <time_3> secs
                And I retrieve a list of remote logistic regression tagged with "<tag>"
                And I create a fusion from a list of models and weights
                And I wait until the fusion is ready less than <time_4> secs
                When I create a prediction for "<data_input>"
                Then the prediction for "<objective>" is "<prediction>"
                And the fusion probability for the prediction is "<probability>"
                And I create a local fusion prediction for "<data_input>"
                Then the local fusion prediction is "<prediction>"
                And the local fusion probability for the prediction is "<probability>"

                Examples:
                | data                | time_1  | time_2 | time_3 | time_4 | data_input | objective | prediction
                | ../data/iris.csv | 10      | 10     | 20     | 20 | {"petal length": 1, "petal width": 1} | "000004" | "Iris-setosa"
        """
        print self.test_scenario6.__doc__
        examples = [
            ['data/iris.csv', '10', '10', '20', '20',
             '{"tags":["my_fusion_6_tag"], "missing_numerics": true}',
             'my_fusion_6_tag',
             '{"petal width": 1.75, "petal length": 2.45}',
             "000004",
             "Iris-setosa",
             '0.4727',
             '{"tags":["my_fusion_6_tag"], "missing_numerics": false, "balance_fields": false }', '[1, 2]']]
        for example in examples:
            print "\nTesting with:\n", example
            source_create.i_upload_a_file(self, example[0])
            source_create.the_source_is_finished(self, example[1])
            dataset_create.i_create_a_dataset(self)
            dataset_create.the_dataset_is_finished_in_less_than(self, example[2])
            model_create.i_create_a_logistic_model_with_objective_and_parms(self, example[8], example[5])
            model_create.the_logistic_model_is_finished_in_less_than(self, example[3])
            model_create.i_create_a_logistic_model_with_objective_and_parms(self, example[8], example[11])
            model_create.the_logistic_model_is_finished_in_less_than(self, example[3])
            compare_pred.i_retrieve_a_list_of_remote_logistic_regressions(self, example[6])
            model_create.i_create_a_fusion_with_weights(self, example[12])
            model_create.the_fusion_is_finished_in_less_than(self, example[3])
            compare_pred.i_create_a_local_fusion(self)
            prediction_create.i_create_a_fusion_prediction(self, example[7])
            prediction_create.the_prediction_is(self, example[8], example[9])
            prediction_create.the_fusion_probability_is(self, example[10])
            compare_pred.i_create_a_local_prediction(self, example[7])
            compare_pred.the_local_prediction_is(self, example[9])
            compare_pred.the_local_probability_is(self, example[10])