pawleymatthew / deepweibull Goto Github PK

Hello @pawleymatthew !

Not sure if you are still support this repo since its been a few years, but I figured heck I'll give it a try...

Firstly, thanks for the great repo. Very well organized code. I didn't have to do too much to update it to new versions of various python packages.

My question in this issue regards reproducing your metabric size experiments.

I reran the full experiment and here are my results:

https://github.com/sourcesync/DeepWeibull/blob/gw/reproduce/plots/real_data_experiments/metabric_training_size.png

This looks a bit different than your results, esp. with respect to the performance of deep_weibull and I'm sure I've missed something.

One thing I noticed is that your repo already captured various experiments from proportion=0.05 to 0.60 in a csv.

Question: Did you run those with special training parameters not captured in the current code?

Anyway, I'm sure I'm missing something obvious.

Thanks for any help you can lend!

( If it helps, my code is in a forked repo as you can see from the experiment results image above. )