Only 0 probabilities for some examples

[hakank]

On some examples, dice shows only 0 probabilities.

• resources/caesar-iter.dice

$ dice caesar-iter.dice
================[ Joint Distribution ]================
Value	Probability	
0	0.	
1	0.	
2	0.	
3	0.	

• benchmarks/simpleCaesar.dice

$  dice simpleCaesar.dice 
================[ Joint Distribution ]================
Value	Probability	
0	0.	
1	0.	
2	0.	
3	0.	
4	0.	
5	0.	
6	0.	
7	0.	
8	0.	
9	0.	
10	0.	
11	0.	
12	0.	
13	0.	
14	0.	
15	0.	
16	0.	
17	0.	
18	0.	
19	0.	
20	0.	
21	0.	
22	0.	
23	0.	
24	0.	
25	0.	
26	0.	
27	0.	
28	0.	
29	0.	
30	0.	
31	0.	

• benchmarks/baselines/digitRecognition.dice

$ dice digitRecognition.dice
================[ Joint Distribution ]================
Value	Probability	
0	0.	
1	0.	
2	0.	
3	0.	
4	0.	
5	0.	
6	0.	
7	0.	
8	0.	
9	0.	
10	0.	
11	0.	
12	0.	
13	0.	
14	0.	
15	0.	

Have I missed something? Should one use some of Dice's flags on these programs?

I'm running Dice on a Linux Ubuntu 18.05LTS,

[SHoltzen]

Thanks so much for this bug report, and thanks for your interest in dice!

This is indeed unexpected behavior which should be handled better. This "all-zeros" output is occurring due to a floating point precision issue (these examples are observing very low probability of evidence).

One of the enhancements we are working on for dice is an arbitrary precision inference mode that can better handle these situations. This is a high priority for us so hopefully this will be implemented soon.

[hakank]

Thanks for your answer. Support for arbitrary precision sounds great.

[loopreal]

Hi,
I notice that the problem is resolved if I build Dice from source. Thanks so much. But the version in the docker image still has all-zeros for these examples. Is it possible to update the docker version?

[rishubn]

Is there any update on this?

[SHoltzen]

Yes, this is resolved now in the main branch. There is a command-line option -wmc-type which permits choosing between different WMC representations (including rational); choosing rational should fix this.