help to explain results

KNIME Analytics Platform
, ,
1

hi everyone.

I don’t know if this is right or there is something wrong here
I’m confused here
scorer 100%, but for the lift curve and roc
It’s confused me
I don’t know why
or what’s wrong
I need anyone to explain this result to me

scorer
scorer561×695 29.1 KB

lift
lift1160×828 68.6 KB

roc
roc1261×948 140 KB

thanks for all

2

What are you trying to do? Could you share your workflow?

2 Likes
3

@mohammad_alqoqa in the last picture you are plotting three prediction against one “truth” (from one class). So the other ones will not match.

Maybe you explain what you want to do and share the workflow side the iris dataset is not confidential.

1 Like
4

svm
svm1029×559 75.9 KB

I use SVM to classify the iris dataset

5

is that because wrong configuration in lift node?

6

Could you please upload the actual workflow rather than a screenshot? Others are more likely to help if they don’t have to manually recreate a workflow. Make sure to store your data files in a “data” subfolder which must bre created inside your workflow folder. Then point the file reader to that folder:

SVM Data Location
SVM Data Location1548×680 88 KB

4 Likes
7

I hope this will be helpful to help me

lft curve test.knwf (58.6 KB)

8

There are several things wrong.

  1. There is no target value column in the scoring data set.
  2. Instead, score the model with the bottom output of the Partitioning Node.
  3. Use the Iris column as the Target
  4. For the ROC curve, pick one species of Iris for the Positive Class Value and include probability for that species, and that will generate an ROC curve. It will be a perfect ROC curve for I. setosa. Curves for the other two species will less so.
  5. for the lift chart, select Prediction (Iris) as the Response Column, and select the probability column for that species and for the label. Do the same thing for the other two species separately.
2 Likes
9

Try this. Its more or less consistent with @Bob_Nisbet 's suggestion + plus writes out the model which can be used to label new data. I’m skeptical about using ROC curves and Lift Charts on such small data sets. They’re not wrong, but don’t provide very useful plots. This workflow is on v. 4.7.7. I’m not currently using 5.2.0 - too buggy.

4 Likes
10

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.