using np.exp() to convert column in pandas DF

Question

I have the following dataframe:

i want to convert the val column to create values that add up to 1

df['val'] = np.exp(df.val)/sum(np.exp(df.val))

but i get the following

id  val
1   0.0
1   NaN
1   0.0

how can i take care of this to make probabilities equal to 1 for this case?

i know it is exploding but if it's just a large number should i just make the value 1? — Eisen, Mar 09 '23 at 19:29
that depends of what you expect from result. What did you mean with "that add up to 1" ? — RomanPerekhrest, Mar 09 '23 at 19:30
This is the softmax function. I'd use some ml package. See the dup question for non-overflow implementation; basically, substract the `val` by its max before applying your function. — Quang Hoang, Mar 09 '23 at 19:33

theprosty · Answer 1 · 2023-03-09T19:32:57.283

0

I hope I understood your question correctly.
You can divide each value in the val column by sum of all the values in the val column:

df['val']/df['val'].sum()

0    0.004792
1    0.964063
2    0.031145
Name: val, dtype: float64

edited Mar 09 '23 at 19:32

answered Mar 09 '23 at 19:32

theprosty

1 Answers1