ContinuousTimeMarkovModel

Implements the model described in the paper

Unsupervised Learning of Disease Progression Models (X. Wang, D. Sontag, F. Wang), KDD'14

http://cs.nyu.edu/~dsontag/papers/WanSonWan_kdd14.pdf

Instructions

• Install the package with the command

pip install --process-dependency-links --trusted-host github.com -e git+https://github.com/evidation-health/ContinuousTimeMarkovModel.git#egg=ContinuousTimeMarkovModel

• cd into the ContinuousTimeMarkovModel directory, something like /usr/local/lib/python2.7/site-packages/src/continuoustimemarkovmodel or try running 'import ContinuousTimeMarkovModel; print ContinuousTimeMarkovModel.path' in a python script

• Add your data using the claimsToInputs.py script in the data file (may need slight modification for specific csv format). Example:

python claimsToInputs.py mydata.csv -c 32 -o myPickledInputs

mydata.csv should contain rows of index, userid, datetime, claimsCode(i.e. ICD9) with a header pers_uniq_id, date_of_service, primary_diag_cd but the script can be easily modified to fit slightly different formatting

• Run the runSontagModel.py script in the examples directory by passing it your inputs (or try out the small_sample default) which in ipython would be run as:

run runSontagModel.py -n 1001 -t 100 -d '../data/myPickledInputs'

• In ipython, plot sampling results using pymc3's traceplot functionality or use groundTruthTraceplot to compare samples to a known result

pymc3.traceplot(trace=trace,vars=['pi']); plt.show()
groundTruthTraceplot(L_start,trace,var='L')