logistic regression - Model training error (H2O)
I am using RapidMiner 9.6.000 and I am having the same issue with the Model training error (H2O) with my logistic regression function. 
I have tried the unchecking compute p-value and it hasn't fixed it. Is there something else I can do in the newest version?
I have tried the unchecking compute p-value and it hasn't fixed it. Is there something else I can do in the newest version?Find more posts tagged with
Sort by:
1 - 4 of
41
Hello @Shaila_Segal
I found the issue. When you are converting the attribute type from numerical to binominal, all the values are set to FALSE, instead of setting 1's to TRUE and 0's to FALSE or vice versa. You can see from the screenshot that all 1000 samples are labeled as FALSE.
The reason for that is the options you selected in the numerical to binominal parameter. The min and max values selected are not correct for this attribute. So, if you selected a min of 0.0 and a max of 1.0 the values between 0.0 and 1.0 (including these two) will be represented as FALSE and others are represented as TRUE. In your case, as you have only 0 or 1 in your label attribute (CHURN), it is converting both 0 and 1 as FALSE. So, to avoid this error you need to select the range as 0.1 and 1.0, this way all the 1's in your label will be converted as FALSE and all 0's will be represented as TRUE.
You should be careful in using "Numerical to binominal", based on your settings it will convert all integer attributes in your dataset based on this condition (>=0.1 and <=1.0) as FALSE and everything else as TRUE.
If you just want to convert the CHURN attribute to binominal you can just select "Single" in attribute filter type and selec CHURN attribute in "attribute" and be sure to change "min" to 0.1 .
I found the issue. When you are converting the attribute type from numerical to binominal, all the values are set to FALSE, instead of setting 1's to TRUE and 0's to FALSE or vice versa. You can see from the screenshot that all 1000 samples are labeled as FALSE.
The reason for that is the options you selected in the numerical to binominal parameter. The min and max values selected are not correct for this attribute. So, if you selected a min of 0.0 and a max of 1.0 the values between 0.0 and 1.0 (including these two) will be represented as FALSE and others are represented as TRUE. In your case, as you have only 0 or 1 in your label attribute (CHURN), it is converting both 0 and 1 as FALSE. So, to avoid this error you need to select the range as 0.1 and 1.0, this way all the 1's in your label will be converted as FALSE and all 0's will be represented as TRUE.
You should be careful in using "Numerical to binominal", based on your settings it will convert all integer attributes in your dataset based on this condition (>=0.1 and <=1.0) as FALSE and everything else as TRUE.
If you just want to convert the CHURN attribute to binominal you can just select "Single" in attribute filter type and selec CHURN attribute in "attribute" and be sure to change "min" to 0.1 .
Sort by:
1 - 1 of
11
Hello @Shaila_Segal
I found the issue. When you are converting the attribute type from numerical to binominal, all the values are set to FALSE, instead of setting 1's to TRUE and 0's to FALSE or vice versa. You can see from the screenshot that all 1000 samples are labeled as FALSE.
The reason for that is the options you selected in the numerical to binominal parameter. The min and max values selected are not correct for this attribute. So, if you selected a min of 0.0 and a max of 1.0 the values between 0.0 and 1.0 (including these two) will be represented as FALSE and others are represented as TRUE. In your case, as you have only 0 or 1 in your label attribute (CHURN), it is converting both 0 and 1 as FALSE. So, to avoid this error you need to select the range as 0.1 and 1.0, this way all the 1's in your label will be converted as FALSE and all 0's will be represented as TRUE.
You should be careful in using "Numerical to binominal", based on your settings it will convert all integer attributes in your dataset based on this condition (>=0.1 and <=1.0) as FALSE and everything else as TRUE.
If you just want to convert the CHURN attribute to binominal you can just select "Single" in attribute filter type and selec CHURN attribute in "attribute" and be sure to change "min" to 0.1 .
I found the issue. When you are converting the attribute type from numerical to binominal, all the values are set to FALSE, instead of setting 1's to TRUE and 0's to FALSE or vice versa. You can see from the screenshot that all 1000 samples are labeled as FALSE.
The reason for that is the options you selected in the numerical to binominal parameter. The min and max values selected are not correct for this attribute. So, if you selected a min of 0.0 and a max of 1.0 the values between 0.0 and 1.0 (including these two) will be represented as FALSE and others are represented as TRUE. In your case, as you have only 0 or 1 in your label attribute (CHURN), it is converting both 0 and 1 as FALSE. So, to avoid this error you need to select the range as 0.1 and 1.0, this way all the 1's in your label will be converted as FALSE and all 0's will be represented as TRUE.
You should be careful in using "Numerical to binominal", based on your settings it will convert all integer attributes in your dataset based on this condition (>=0.1 and <=1.0) as FALSE and everything else as TRUE.
If you just want to convert the CHURN attribute to binominal you can just select "Single" in attribute filter type and selec CHURN attribute in "attribute" and be sure to change "min" to 0.1 .
Can you check the input going into a logistic regression operator? The error represents that there is only one label going inside the logistic regression operator. As prediction cannot be done with one label it might be throwing an error.
You can check the input by right-clicking on the logistic regression operator and select "breakpoint before". Once you run the process you can see the data going into the operator and in statistics you can see the label statistics representing the number of samples in each class.
If you want us to check, please provide us with the process by going to FILE --> Export process and attach that .rmp file here. Please also provide us with data.