0
$\begingroup$

I have a linear tabular dataset made of floats. The dataset follows a simple rule like:

if features A and B are in a certain range then target class is 1, otherwise target class is 0.

Since I want to get some interpretability from my ANN model, I opted for using the integrated gradients method implemented by alibi.

Unfortunately, most of individual samples don't show A and B as the leading features as expected. Even more weird is the fact that, when I average the attributions of all the individual samples, A and B get the highest score. In other words, local explanations fail but, on average, the global explanation is correct.

Can anyone help me out to understand why this happens? Isn't integrated gradients method suitable for tabular datasets?

By the way, my baseline is based on a uniform distribution of random floats ranging from 0 to the maximum of each column.

$\endgroup$
0
$\begingroup$

After several tests, I've opted for the following configuration:

Approximation mode: riemann trapezoidal Steps: 1000 Baseline: average mean of each column (feature)

Although the global score remains good, there're important variations in the number of correctly explained local samples. For example, an execution returns 200 out of 390 samples explained as expected, but the following execution only 17.

Is there a reason for that? And, is there a way to make the algorithm predictable?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.