The branching model has non-deterministic error

[xukai92]

The branching model in the branching_aistats.clj file has some non-deterministic error.

(defn fib [n]
  "returns the n-th number in the Fibonacci sequence"
  (loop [a 0 b 1 m 0]
    (if (= m n)
      a
      (recur b (+ a b) (inc m)))))

(with-primitive-procedures [fib]
  (defquery branching
      "A simple example illustrating flow control with
      dependence on random choices"
      []
      (let [count-prior (poisson 4)
            r (sample count-prior)
            l (if (< 4 r)
                6
                (+ (fib (* 3 r))
                   (sample count-prior)))]
        (observe (poisson l) 6)
        (predict :r r)
        (predict :l l))))

If I run it with a lot of particles with SMC sampler, it sometimes give Error during inference: org.apache.commons.math3.exception.NotStrictlyPositiveException: mean (0) (but sometimes not). If the # of particles is large (say 500,000), it will be broken nearly every time.

[fwood]

Good catch!

Many apologies both for the delay and for the error.

Please note that this is a simple bug in the implementation of fib. The
correct code can be found here
http://www.robots.ox.ac.uk/~fwood/anglican/examples/viewer/?worksheet=aistats/branching-aistats.
For ease of reference I include it here

(defn fib [n]
"returns the n-th number in the Fibonacci sequence"
(loop [a 1 b 1 m 0](if %28= m n%29
a
%28recur b %28+ a b%29 %28inc m%29%29)))

---

Dr. Frank Wood
Associate Professor - Information Engineering, Oxford
Fellow - Kellogg College, Oxford
Fellow - Alan Turing Institute
Telephone: +441865283060
Web: http://www.robots.ox.ac.uk/~fwood
Email: [email protected]
Office: IEB 20.03

University of Oxford
Department of Engineering Science
Parks Road
Oxford
OX1 3PJ

Assistant: Maddy Crudge
Email: [email protected]
+44 1865 283058

Where: http://www.eng.ox.ac.uk/contact/how-to-reach-the-department

On Tue, Nov 8, 2016 at 9:03 PM, Kai Xu [email protected] wrote:

The branching model in the branching_aistats.clj file has some
non-deterministic error.

(defn fib [n]
"returns the n-th number in the Fibonacci sequence"
(loop [a 0 b 1 m 0](if %28= m n%29
a
%28recur b %28+ a b%29 %28inc m%29%29)))

(with-primitive-procedures [fib](defquery branching
"A simple example illustrating flow control with
dependence on random choices"
[]
%28let [count-prior %28poisson 4)
r (sample count-prior)
l (if (< 4 r)
6
(+ (fib (* 3 r))
(sample count-prior)))]
(observe (poisson l) 6)
(predict :r r)
(predict :l l))))

If I run it with a lot of particles with SMC sampler, it sometimes give Error
during inference: org.apache.commons.math3.exception.NotStrictlyPositiveException:
mean (0) (but sometimes not). If the # of particles is large (say
500,000), it will be broken nearly every time.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1, or mute the
thread
https://github.com/notifications/unsubscribe-auth/AADZnkNFhQaQ2O7wrv3sBV2o1LZBSkm8ks5q8OOGgaJpZM4Ks54l
.