sabazm/counterfactual-prediction

State space model + data pipeline to generate counterfactual time series trajectories on multiple clinical signals, used to evaluate the utility of counterfactual features in sepsis prediction

Counterfactual-Prediction

Create counterfactual trajectory of a signal given past trajectory and some contexts (Please see the pdf file for a more detailed description of the model and the data pipeline)

Files Description

• PDF file
  • Detailed write up of the model and results
• Python files
  • EM.py: perform inference and parameter estimation
  • plot.py: plot predicted states and observations
• Notebook files
  • Pipeline_Preprocessing: extract relevant data from database files; preprocess; perform inference, parameter estimation and prediction
  • Pipeline_Training: given preprocessed file, perform training, can perform hyperparameter validation and training on multiple signals through parallel computing (using ipyparallel)
  • DLM: simulation to evaluate model's performance
• Old dev files
  • EM_individual_params: EM for learning individual parameters (incomplete)
  • preprocess.py: preprocess data into format usable for EM.py (more general version is in Pipeline_Preprocessing.py)
  • Population Level Analysis: assume the same set of parameters for all patients; fit to real data
  • Individual Level Analysis: assume a set of parameters for each patients; fit to real data
  • Source Data Analysis: some analysis on the real data set before and after preprocessing
  • Playground: place to do small analysis and testing