mlwave/kaggle-ensemble-guide

Code for the Kaggle Ensembling Guide Article on MLWave

kaggle-ensemble-guide's Issues

Geomean

Sometimes geomean (i.e https://en.wikipedia.org/wiki/Geometric_mean) gives us a better score than average for some particular reasons.
Do you think it's worth it adding kaggle_geomean.py?

Add weightage parameters for more customized blending.

I'm still thinking of a way to do it elegantly, maybe something like this:

$ cat files.csv
./samples/method1.csv,0.5
./samples/method2.csv,0.25
./samples/method2.csv,0.25

$ python kaggle_rankavg.py files.csv "./samples/kaggle_rankavg.csv"
parsing: ./samples/method1.csv with weight: 0.5
parsing: ./samples/method2.csv with weight: 0.25
parsing: ./samples/method3.csv with weight: 0.25
wrote to ./samples/kaggle_rankavg.csv

What do you think?

Submissions will need to be sorted before ensembled.

Often time, submissions can come from different language / libraries that might affect their orders.
I think we should have an option (or by default) to sort them beforehand to make sure we ensemble the right ids together 😃

Have a version of blend_proba.py for python 3.5

Hi,

can you write a version of blend_proba.py for python 3.5,

Please see: http://stackoverflow.com/questions/39687959/error-when-converting-form-python-2-to-python-3

Regards,

How do you use glob files

for the first argument, how do you use glob files?

The link is gone. Is there any other Internet address?

http://mlwave.com/kaggle-ensembling-guide/

Error in kaggle_vote weighted

Executing

$ python kaggle_vote.py "./samples/_*.csv" "./samples/kaggle_vote.csv" "weighted"

Is giving weird result / output.