Skip to content

iliang1234/CS-282-Final-Project

Repository files navigation

CS-282-Final-Project

Code repository for the IRL What-If Tool (IRL-WIT), an interactive visualization tool designed to enhance interpretability in Inverse Reinforcement Learning (IRL), particularly in the healthcare domain.

Directory Description

/AIRL

Attempt at implement AIRL (in progress)

/Experiment1_PCA

(Checkpoint 3 output viz.ipynb) Notebook code for PCA and output feature data analysis.

/Experiment2_reward_landscapes

Adjusted code to create multi-dimensional plots that explain how reward values grow during the training process

  • Paper citation: Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer and Tom Goldstein. Visualizing the Loss Landscape of Neural Nets. NIPS, 2018.

/Experiment3_gridworld

(extract_gridworld.ipynb) Notebook to extract reward data for the model running on gridworld toy domain.

/Experiment4_data

(hypotension_extractions_clusters.ipynb) Notebook to extract rewards for the model running on hypotension patient cohort.

/Experiment5_data

(pneumonia_extractions_clusters.ipynb) Notebook to extract rewards for the model running on pneumonia patient cohort, for one clustering algorithm.

(pneumonia_clustering.ipynb) Notebook to extract reward value information for each clustering algorithm.

/data_pipeline

Overall purpose here is to extract the pneumonia patient cohort.

(preproc_mapping.py) Downloads emars table to map (subject_id, admission_id) —> list of drugs/treatments given to patient

  • The mapping is stored in treatments.json (todo: rename the file to a more descriptive title)

(pneumonia_treatments.py) Reads in the mapping above and identifies the unique set of treatments given to pneumonia patients in the past

(pneumonia_cut.py) Exploratory code to append an action to each patient’s row in data frame and export final data frame

/IRLalgorithms_code

Compilation of extracted reward / gradient values and code for optimizer, trajectory construction etc.

/reward landscape

(BC_loss_landscape.ipynb) Adjusted code to create multi-dimensional plots that explain how reward values grow during the training process

  • Paper citation: Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer and Tom Goldstein. Visualizing the Loss Landscape of Neural Nets. NIPS, 2018.

/WIT_framework

(CS282_Final_Project_Pneumonia_Viz.ipynb) IRL-WIT code for pneumonia cohort (CS282_Final_Project_MIMIC_IV_viz.ipynb) IRL-WIT code for hypotension cohort

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published