Generate interpretation result

INHYEOK_SONG

When I specify this kernelSHAP, I am wondering what value of interpretatioin is made and what it means.

MartinLiebig

Hi,

those algorithms usually answer the question:

What would be the result of the model, if the value for this variable (and sometimes the other variables) is not what it is, but distributed like the rest of the data?

So for your first row, If a2 changes, then your prediction will change in average by 38.647. The difference between the algorithms is how to define a 'change'.

Best,

Martin

MartinLiebig

Hi,

those algorithms usually answer the question:

What would be the result of the model, if the value for this variable (and sometimes the other variables) is not what it is, but distributed like the rest of the data?

So for your first row, If a2 changes, then your prediction will change in average by 38.647. The difference between the algorithms is how to define a 'change'.

Best,

Martin

INHYEOK_SONG

Thank you. I'll ask you one more question. I am wondering why there is only 3 numbers appearing in the interpretaion column. I am wondering if there are any standards displayed.

MartinLiebig

Hi,

currently this column only has the top 3. We may add an option to get more there.

The full list of influence factors is available at the 'imp' port of the operator if you need them.

Best,

Martin