Prev: plac 0.5 is out!
Next: Pick items from list with probability based upon property oflist member ?
From: southof40 on 21 Jun 2010 06:01
On Jun 20, 11:27 pm, Mel <mwil...(a)the-wire.com> wrote:
> southof40 wrote:
> > I have list of of N Vehicle objects - the only possible vehicles are
> > cars, bikes, trucks.
>
> > I want to select an object from the list with a probability of : cars
> > 0.7, bikes 0.3, trucks 0.1.
>
> > I've currently implemented this by creating another list in which each
> > car object from the original list appears 7 times, each bike 3 times
> > and each truck once. I then pick at random from that list.
>
> This seems about right.  It's like a lottery where the more likely
> winners have more tickets, but all tickets are the same.  Pick one to
> pick the winner.
>
> There's a very expensive, but flexible technique that effectively gives
> some tickets a better chance than others.  You have to examine each
> ticket individually, so this algorithm burns random numbers like
> kindling:
>
> import random
>
> #===============================
> def weighted_choice (candidates, weight):
>     chosen = None
>     total = 0
>     for c in candidates:
>         w = weight (c)
>         total += w
>         if random.randint (0, total-1) < w:
>             chosen = c
>     return chosen
> #===============================
>
> def test_weight (c):
>     return {'c':7, 'b':3, 't':1}[c]
>
> def item_count (s, target):
>     return sum (1 for x in s if x==target)
>
> test_candidates = 'c'*100 + 'b'*100 + 't'*100
>
> for i in xrange (10):
>     test = [weighted_choice (test_candidates, test_weight) for k in xrange (100)]
>     for x in 'cbt':
>         print x, item_count (test, x), '\t',
>     print
>
>         Mel.

I like this - makes altering the probabilities straightforward and
this is something I may want to do.
From: southof40 on 21 Jun 2010 06:04
On Jun 20, 10:58 pm, Cameron Simpson <c...(a)zip.com.au> wrote:
> On 20Jun2010 12:44, Stefan Behnel <stefan...(a)behnel.de> wrote:
> | southof40, 20.06.2010 12:19:
> | >I have list of of N Vehicle objects - the only possible vehicles are
> | >cars, bikes, trucks.
> | >
> | >I want to select an object from the list with a probability of : cars
> | >0.7, bikes 0.3, trucks 0.1.
> | >
> | >I've currently implemented this by creating another list in which each
> | >car object from the original list appears 7 times, each bike 3 times
> | >and each truck once. I then pick at random from that list.
> | >
> | >This works but seems very clunky to me.
> |
> | Why? It's a very simple, generic, easy to understand and fast
> | solution to the problem.
>
> Only 3 out of 4, if you want to be precise in your selections.
> Supposing he wants probabilities 0.7432, 0.3765, 0.1087654 ?
> The required list needs to be Very Long to achieve an accurate
> representation, and thus Very Slow to construct/populate.
>

Yes you're spot on here. Although I have used simple probabilities it
occurred to me that the list for my current method would get pretty
big if I changed the probs to be a little more refined.

First  |  Prev  | 
Pages: 1 2
Prev: plac 0.5 is out!
Next: Pick items from list with probability based upon property oflist member ?