ml-jku / hopular Goto Github PK

I'm reading the dataset code and there is probably a bug here:

After that, you are using the old indices to index into the shuffled/concatenated arrays. So the splits are different (not stratified, for example):

It repeatedly says that the ModelCheckpoint is not configured to save the best model. Please help to fix this issue

HI ml-jku
I am trying to run hopular on my own data sets. It looks like there are two formats for doing this

• one, used for blastchar, just involves adding a csv file and describing it in auxiliary/data.py. I have done this for the "adult census" data and it all seems to work (ie. no errors)

• the other seems to involve the use of about 4 files i.e for statlog_heart it is

  • folds_py.dat
  • labels_py.dat
  • statlog_heart_py.dat
  • validation_folds_py.dat

• I can't make any sense of this. The heart data should have 6 numeric and 7 discrete features, according to both data.py and the UCI website. however. it seems to have 13 numeric features

> head -2  resources/statlog_heart/statlog_heart_py.dat 
1.70892,0.688222,0.869313,-0.0752701,1.39961,-0.416256,0.979844,-1.75595,-0.699923,1.17882,0.675165,2.4681,-0.874083
1.37958,-1.44764,-0.183219,-0.91506,6.08171,-0.416256,0.979844,0.445582,-0.699923,0.480261,0.675165,-0.710216,1.18707

also I cant figure out what folds_py.dat and validation_folds_py.dat are doing.

> head -2  resources/statlog_heart/folds_py.dat 
0,1,0,0
1,0,0,0

> head -2  resources/statlog_heart/validation_folds_py.dat 
1,0,1,0
0,0,0,0

I am not so interested in the heart data as in understanding how to add my own data, use a test set or cross-validation and then
test on a independent data set.

Bye

@bschaefl Hi, hopfield networks are very interesting and understudied indeed however, can you please answer this?
https://medium.com/@tunguz/trouble-with-hopular-6649f22fa2d3

Hello
LightGBM/CatBoost/XGBoost offer the ability to play with non rescaled features with plenty of NaN in the dataset. This is specialy true when theses tools are invoked in FLAML which greatly simplifies their usage.
Do you plan to allow NaN in any cell from any column ? What is the impact of non rescaling data ?
Thanks

I looked into the repository and did not find an easy way to use our code in my own datasets. Something I would expect is an interface similar to the regression functions on scikit learn. This would be amazing for other researchers and students to use this new network. Is there any script, change or material available to help use this repository this way?

E.g. hopular.fit(x,y)

Thanks!