new operator simm add
Just briefly introducing a new operator. Not sure how useful it is going to be, but may as well document it. Basically, I wanted to implement this equation:
R0 = r0
R_k = R_k-1 + r_k * [simm(R_k-1,r_k)]^p
where r_k are superpositions, simm is our similarity metric, and if we set p = 0 this reduces to a simple sum:
R_k = r1 + r2 + ... + r_k
The intended effect is that you add new superpositions weighted somewhat on how close they are to those already in the sum. Noting of course, that simm gives results in [0,1]. Like I said, I don't know if this will turn out useful, but figured why not add it.
Here is the python:
op,p = parameters.split(',')
p = int(p) # maybe try float(p) later
head, *tail = one
Rk = head.apply_op(context,op)
for x in tail:
rk = x.apply_op(context,op)
similarity = silent_simm(Rk,rk)**p
Rk = Rk + rk.multiply(similarity)
And a use case from earlier:
sadd |result 1> => simm-add[noise,1] select[1,1] rel-kets[noise] |>
sadd |result 2> => simm-add[noise,1] select[1,2] rel-kets[noise] |>
sadd |result 40> => simm-add[noise,1] select[1,40] rel-kets[noise] |>
Though in that case the difference from just a plain sum was essentially zero.
In general, usage is simply: simm-add[op,p] with some similarity to common[op] and union[op].
May as well note, since simm is in [0,1], and assuming all superpositions have positive coeffs, then we have the property:
0 <= R_k <= r1 + r2 + ... + r_k
where it reaches the upper bound only if p = 0, or all r_k are identical. The more "diverse" the superpositions (ie, smaller simm values), the smaller R_k will be.
previous: averaging out noise
next: introducing a new tool wikivec and wikivec similarity
by Garry Morrison
email: garry -at- semantic-db.org