Advanced Functionalities

The guide provided on Getting Started page uses the wrapper function hardy_main which sequentially executes data_wrapper, classifier_wrapper and report generation. The structure of HARDy package is shown in the image below:

image explaining the structure of HARDy

data_wrapper parses through the transformation configuration file and loads it into the environment. The data wrapper also loads the .csv files into the environment. The data wrapper then applies the transformations, outlined in the configuration file, to the data files. The data_wrapper stores the tranformed information into the .pkl file or into .png files depending on the user defined arguments.

To save time data_wrapper is capable of parallelizing the transformations process. The parallelization is controlled through the n_threads parameter. By default, n_threads is set a 1.

classifier_wrapper loads the data generated by data_wrapper. By default, classifier_wrapper deletes the .pkl generated by the data_wrapper. The loaded data is then ran through the convolutional neural network (CNN) or hyperparameter tuning sessions depending upon the specified inputs from the user.

Each tuning session is then reported into data_reporting module. The tuned/trained model is then validated againt num_test_files_class. Then the report folder having project_name in data_path is created. The report folder contains folders for each transformation run. These transformation folders contains model validation results, best tuned model for a particular transformation and model hyperparameter details.