Predicting ints on titanic dataset.

Excuse my noob level of understanding please, I'm brand new.

I am trying to predict mortality on the titanic dataset using cross-validation and linear regression. As you can only use numbers with linear regression, I have converted selected attributes (such as survived) using the 'nominal to numerical' operator. I can see it is working most of the time from looking at the data and rounding it to 1 or 0 however the predicted value is coming back as a double so it's showing as 0 correct predictions.

I suppose my question is how do I make rapidminer return an int instead of a double? I have tried using the 'real to integer' operator but it doesn't like me putting it anywhere!

Open to any suggestions.

Image: https://us.v-cdn.net/6030995/uploads/editor/m7/ruy87i7a1pi2.png

Find more posts tagged with

AI Studio

Cross Validation

Apply Model

Nominal to Numerical

Linear Regression

Accepted answers

MartinLiebig

Hi,

First, if you use a GLM operator it can handle binominal data. it uses the same trick you are doing here but without any hazzle for you.

Then, why exactly do you want a int over a double?

Anyway, one way to do it is to use Generate Attributes with

round([prediction(Survived=Yes)])

Best,

Martin

All comments

VenomSwitch

When I say 'double' I actually mean 'real'.

MartinLiebig

Hi,

First, if you use a GLM operator it can handle binominal data. it uses the same trick you are doing here but without any hazzle for you.

Then, why exactly do you want a int over a double?

Anyway, one way to do it is to use Generate Attributes with

round([prediction(Survived=Yes)])

Best,

Martin

VenomSwitch

Hi Martin,

This is brilliant, GLM is exactly what I needed!

I wanted the integer because it was giving me values with the decimal point and becuase they didn't exactly match the '1'/'0' in the survived column it just told me every one was wrong with 0% accuracy (as it wasn't rounded to the '1'/'0' format in the dataset).

I couldn't figure out where to place the generate attributes operator but it doesn't really matter as GLM has sorted out my problem.

A very handy operator.

Cheers!

Joel

MartinLiebig

Hi @VenomSwitch ,

you would basically put it after each and every apply model operator you are using. Great that the GLM worked.

Best,

Martin

VenomSwitch

@mschmitz

I have got it working but now it seems to have a 100% accuracy rate which seems suspicious. I'm just going to stick with the GLM process I had before I think! If it isn't broke, don't fix it haha.

You've helped me out today though!

MartinLiebig

Are you sure you applied the round on the prediction and not on the label attribute? That would explain it

VenomSwitch

Here is my current process using generate attributes with linear regress instead of GLM.

My label is 'Survived = Yes'.

I tried using the same operator inside the cross-val aswell but same result; 100% correct prediction.

Image: https://us.v-cdn.net/6030995/uploads/editor/b2/y88zwrksk0wd.png

MartinLiebig

Hi @VenomSwitch ,

then the other idea, are you sure that Survived = No is not part of the training ? That would also explain good results

cheers,

Martin