Repository of Deep Propensity Network - Sparse Autoencoder (DPN-SA) to calculate propensity score using sparse autoencoder - Paper under review at Journal of the American Medical Informatics Association.
A deep learning model - deep propensity network using a sparse autoencoder (DPN-SA), for calculating Propensity Score Mathcing(PSN) to tackle the problems of high dimensionality and residual confounding. It uses a sparse autoencoder in place the Propensity dropout module in Deep counterfactual network - Propensity dropout architecture(DCN-PD).
The original paper of DCN-PD can be founed here.
The original implemetation of DCN-PD in pytorch can be found here.
pytorch - 1.3.1
numpy - 1.17.2
pandas - 0.25.1
scikit - 0.21.3
matplotlib: 3.1.1
python - 3.7.4
causal AI, biomedical informatics, deep learning, multitask learning, sparse autoencoder
Update the DCN Model path
To reproduce the experiments mentioned in the paper for DCN-PD, Logistic regression, Logistic regression Lasso and all the SAE variants of 25-20-10- (greedy and end to end) for both the original and synthetic dataset, type the following command:
python3 main_propensity_dropout.py
After the run, the outputs will be generated in the following location:
Consolidated results will be available in textfile in /Details_original.txt and /Details_augmented.txt files.
The details of each run will be avalable in csv files in the following locations:
/MSE/Results_consolidated.csv and /MSE_Augmented/Results_consolidated_NN.csv
© DISL, University of Florida
Licensed under the MIT License