fixed 'os' import in the 'bulkDGD.core.model' module

setup.py updated with newest version

Merge pull request #13 from Center-for-Health-Data-Science/update_rec…

…ount3

Several updates to the Recount3 subpackage

committed stray bulkDGD/util.py file

Merge pull request #12 from Center-for-Health-Data-Science/fix11_expe…

…riment_attributes

recount3 sub-package updated

### Added

* The 'train()' method was added to the 'core.model.DGDModel' class to train the DGD model.

* The 'dgd_train' executable was added to train the DGD model using the command line.

* A new type of configuration file containing the options to train the DGD model is available. An example of such a configuration file can be found in the newly created 'bulkDGD/ioutil/configs/training' directory inside the package. This file, together with the other configuration files, is installed together with the package.

* A new example of configuration file ('model_untrained.yaml') containing the options to set up the DGD model is available in the 'bulkDGD/ioutil/configs/model' directory for when the model needs to be set up before training.

* A new tutorial on how to train the DGD model was added to the documentation (Tutorial 3).

* The 'load_loss()' and 'save_loss()' functions were introduced in the new 'bulkDGD.ioutil.lossio' module to load and save CSV files containing the losses reported during the training procedure.

### API-breaking changes

* The configuration file used to find the representations for new samples now has a new format. Please refer to the documentation for a detailed explanation of the new format. The format is not backward compatible.

### Internal changes (for contributors)

* The '_get_data_loader()' method has been introduced into the 'core.model.DGDModel' class to create data loaders.

* There is now only one internal method in the 'core.model.DGDModel' class responsible for optimizing one or multiple representations for a set of samples, '_optimize_rep()'. The '_get_representations_one_opt()' and the '_get_representations_two_opt()' methods have been updated accordingly.

* New sanity checks have been introduced when loading configurations ('ioutil.configio' module).

* A new '_get_final_dataframes_train' method has been introduced to create the data frames produced by the new 'train()' method in the 'core.model.DGDModel' class.