Pages: [1]
  Print  
Author Topic: [SOLVED] Plotting ROC curves from separate, saved performance vectors  (Read 633 times)
x2xx7x
Newbie
*
Posts: 6


« on: November 10, 2013, 07:45:18 PM »

Hello everyone,

I have trained several classification models using 5-fold X-Validation, some of which required quite some time to finish and I have saved all of the models and their respective performance vectors. I was wondering if there is any way to merge these performance vectors so that I can plot the ROC curves from each performance vector on the same plot for comparison. I am aware of the Compare ROC function however I do not want to re-train the models on the data.

Will placing the saved models in the Compare ROC function provide the correct output?
I was able to get an output by connecting a data set to the compare ROC and placing the saved models in the Compare ROC, with the model out connected directly to the compare ROC model outs. I am not sure if this is providing the result I am looking for

Thank you,

-Alex
« Last Edit: November 11, 2013, 04:24:45 PM by x2xx7x » Logged
Marius
Administrator
Hero Member
*****
Posts: 1794



WWW
« Reply #1 on: November 11, 2013, 03:18:57 PM »

Hi Alex,

using the stored models in Compare ROCs will deliver correct results, however usually the operator internally performs a cross validation to produce confidence intervals etc. on the ROC curves. This will of course not work if you use the precalculated models. If you can do without this info you approach is fine.

Best regards,
Marius
Logged

Please add [SOLVED] to the topic title when your problem has been solved! (do so by editing the first post in the thread and modifying the title)
Please click here before posting.
x2xx7x
Newbie
*
Posts: 6


« Reply #2 on: November 11, 2013, 03:57:09 PM »

Is there any other way to do this? Maybe somehow to use the already calculated performance vectors with ROC curves and combine the, on one graph. 

If not, thank you.
Logged
Marius
Administrator
Hero Member
*****
Posts: 1794



WWW
« Reply #3 on: November 11, 2013, 04:06:50 PM »

No, unfortunately that's not possible.

Best regards,
Marius
Logged

Please add [SOLVED] to the topic title when your problem has been solved! (do so by editing the first post in the thread and modifying the title)
Please click here before posting.
Pages: [1]
  Print  
 
Jump to: