Currently in test_api, we are using the runkey.config_id as the num_run for that model. However, the config_id in runkey comes from configuration.config_id of the pipeline configuration, i.e, does not take into account dummy or traditional classifiers which the num_run does. So, num_run = C + config_id where C is the number of dummy or traditional classifiers run.

This bug is causing the tests to fail for many PRs.