Add usage example notebooks

[chiragnagpal]

trafficstars

Currently we do not have notebooks that compare performance of DSM with other models.

We would need to compare against: -DeepHit -DeepSurv -Random Survival Forest -Cox PH on Time Dependent CI and Brier Score.

[kishanmaharaj]

@chiragnagpal sir, This looks interesting... I would like to work on this part...

[chiragnagpal]

@kishanmaharaj great, I am assigning this to you. I'll add a small notebook example and let you work on improving this. Ideally we'd want to work on a separate branch for this and not commit to master.

[chiragnagpal]

I have added an Ipython notebook to the branch https://github.com/chiragnagpal/DeepSurvivalMachines/tree/usage_examples that trains and tests DSM in a 5 fold cross validation fashion and computes the Brier Score and Concordance Index:

https://github.com/chiragnagpal/DeepSurvivalMachines/blob/usage_examples/examples/SUPPORT_example.ipynb Next Steps:

1. Add AUC(t) at the 25th, 50th and 75th event quantile.
2. Compare to CoxPH, RSF and Weibull AFT
3. Show how adding non linearities (Hidden layers of MLP) improves model perfromace.
4. Bootstrap the results to generate confidence intervals around the metrics (This can be made a utility function in the library)
5. Proper documentation with appropriate inline latex and markdown.

For hints on the kind of notebook I am talking about https://github.com/IBM/causallib/blob/master/examples/hemm_demo.ipynb

[kishanmaharaj]

@chiragnagpal Okay Sir... Understood.

[kishanmaharaj]

@chiragnagpal sir, can you send me some references on the above-mentioned models (DeepHit, DeepSurv, Random Survival Forest etc.)...?

[chiragnagpal]

you can find the original references to the papers/methods mentioned in our paper https://arxiv.org/pdf/2003.01176.pdf see sec. 4.2

Fpor experiments, you might want to check out the python package pycox, scikit-survival and lifelines which have implementations of these methods.

[kishanmaharaj]

@chiragnagpal Okay, sir...