I reckon the answer is 23. Here's my reasoning.
Consider the partition of P into sets as follows:
So, for example, , etc
Now if we choose an to be in our subset , then we must not choose an , for then
Thus the subset can contain at most element from elements from , and then elements from any two of the remaining , in such a way that no two of the 's add up to .