Multitask Hurdle Causal Neural Networks

Submission of the seminar paper (jupyter notebook) for the Applied Predictive Analytics seminar

The project's main goal was to implement a multitask hurdle causal network that would be able to handle classical hurdle models while dealing with the causal aspect in the data.

Folder structure:

.
├── scripts                              # Implementations of the NN models, tuning func, etc.
│   ├── baseline_nns.py                  # Baseline simple NNs       
│   ├── dataloaders.py                   # PyTorch Dataloaders
│   ├── helper_funcs.py                  # Functions for calculation of the analytical targeting policy and the transformed
│   │                                    # outcome loss, seed function
│   ├── mt_residual_nns.py               # Multitask Hurdle Residual NNs
│   ├── mt_residual_nns.py               # Multitask Hurdle NNs   
│   └── tuning.py                        # Functions for tuning hyperparams of the models
│
├── tuning_results                       # All files needed for running the notebook as is
│   ├── daw_tune                         
│   │   ├── models                       # Saved model checkpoints of the best models tuned with the 
│   │   │                                # Dynamic Weight Average approach
│   │   ├── training_history             # CSV files with training histories for all hyperparameter combinations
│   │   └── daw_tune_results.npy         # nparray with training results of the Dynamic Weight Average approach
│   ├── hyperparam_tune                  # Contains files relevant for MTNet, MTXNet and MTCat
│   │   └── ...                          # Similar to daw_tune
│   ├── residual_model                   # Contains files relevant for the residual models
│   │   └── ...                          # Similar to daw_tune
│   ├── separate_lr_tune                 # Contains files relevant for MTNets trained with multiple learning rates
│   │   └── ...                          # Similar to daw_tune
│   └── simple_nn_tune                   
│       ├── checkout_nn                  # Saved CheckoutNN model checkpoints
│       ├── conversion_nn                # Saved ConversionNN model checkpoints
│       ├── checkout_nn_results.npy      # nparray with CheckoutNN training results
│       └── conversion_nn_results.npy    # nparray with ConversionNN training results
│
├── data                                 # Dataset Folder
│
├── submission.ipynb                     # Main Notebook detailing the project and its results
└── README.md                            # Readme file