MENU
|
|
|
You can contribute this section with improved models or new ones. Please send them to Manfred Jaeger
Models marked "Leuven exercise" are contributed by Bruynooghe et (many!) al. from Leuven University. See the ILP 2009 paper "An Exercise with Statistical Relational Learning Systems".
| Model | Comments | Example queries |
| bloodtype.blp | Reference model. blood_founder.blp in Balios distribution (one cpt row corrected) . | P(bloodtype(linus))=[a:0.32,b:0.31,ab:0.20,null:0.16]
P(bloodtype(linus)|pchrom(uwe)=a,mchrom(uwe)=a)=[a:0.4391,b:0.2092,ab:0.2354,null:0.1163]
(approximate values: likelihood sampling with sample size 50000) |
| bloodtype.mln bloodtype.db |
Bloodtype encoding for Alchemy. |
Some initial Results, and
Feedback from Alchemy developer team. |
| bloodtype.psm | Bloodtype encoding for Prism. | P(bloodtype(linus))=[0.32,0.32,0.20,0.16] (exact inference) |
bloodtype.rbn
bloodtype.rst |
Bloodtype encoding for Primula. |
P(bloodtypeA(linus))=0.3218
P(bloodtypeB(linus))=0.3208
P(bloodtypeAB(linus))=0.1978
P(bloodtype0(linus))=0.1594
P(bloodtypeA(linus)|mchromA(uwe),pchromA(uwe))=0.4386
P(bloodtypeB(linus)|mchromA(uwe),pchromA(uwe))=0.2130
P(bloodtypeAB(linus)|mchromA(uwe),pchromA(uwe))=0.2365
P(bloodtype0(linus)|mchromA(uwe),pchromA(uwe))=0.1117
(exact inference; numerical inaccuracies due to rounding in binarization of multi-valued predicates)
|
In order to create relational structures for this domain we have developed a Pedigree generator. This program has different options which can be set by command-line parameters or by a configuration file. In that archive you can find the executable program, some configuration files and the output for the different systems and even in RDEF.
| Model | Comments | Example queries |
| university.mln | Reference model. |
P(advisedBy(Gail,Glen))=?
P(inPhase(Hanna,Pre_Quals))=?
P(inPhase(Hanna,Post_Quals))=?
More detailed Results
and
Feedback from Alchemy developer team. |
university.blp
university-blp-evidence.txt |
University encoding for Balios. Model must be conditioned on evidence in university-blp-evidence.txt |
P(advisedBy(Gail,Glen))=?
P(inPhase(Hanna,Pre_Quals))=?
P(inPhase(Hanna,Post_Quals))=?
Details
|
university.rbn
university.rst | University encoding for Primula. Model must be conditioned on the same evidence as the blp model.
Evidence must be entered through the Primula inference module (no read-in from file possible) |
P(advisedBy(Gail,Glen))=0.11
P(inPhase(Hanna,Pre_Quals))=0.1675
P(inPhase(Hanna,Post_Quals))=0.8297
Details
|
| Model | Comments | Example queries |
|
multistatehmm8.blp
multistatehmm16.blp
multistatehmm32.blp
multistatehmm64.blp
|
A simple HMM model for a random walk on a line with 8,16,32,64 positions
(= number of hidden states), and a 2-state observable |
8 states: P(hiddenstate(10)|position(5)=left)=
[s1:0.21,s2:0.20,s3:0.19,s4:0.15,s5:0.11,s6:0.08,s7:0.05,s8:0.03] |
|
multistatehmm8.psm
multistatehmm16.psm
multistatehmm32.psm
multistatehmm64.psm
|
|
|
|
multistatehmm8.rbn
multistatehmm16.rbn
multistatehmm32.rbn
multistatehmm64.rbn
|
|
8 states: P(hiddenstate(10)|position(5)=left)=
[s1:0.2204,s2:0.2040,s3:0.1763,s4:0.1424,s5:0.1074,s6:0.0737,s7:0.046,s8:0.0298]
|
| Model | Comments |
Example queries |
noisy-or.rbn
large.rst,small.rst
|
Reference model. Very simple noisy-or problem for Primula in two scenarios, large and small.
|
P(on(3))=0.81
P(on(3)|!on(0))=0.67 (small scenario)
|
| noisy-or_large.psm
|
Noisy-or encoding for Prism with the large scenario.
|
| noisy-or_large_sample.psm
|
Noisy-or encoding for Prism with the large scenario with a sampling facility for approximate inference.
|
| noisy-or_small.psm
| Noisy-or encoding for Prism with the small scenario. |
P(on(3))=0.81 |
| noisy-or_small.blp
| Noisy-or encoding for Balios with the small scenario. |
P(on(3))=0.81
P(on(3)|!on(0))=0.67 |
|
- Real Estate (Leuven Exercise)
| Model | Comments | Example queries |
realestate.mln
realestate_example.db | MLN encoding for real estate and one possible database | |
| realestate.blp | BLP for the `real-estate' entity-relationship model. The hard constraint that each house is bought by at most one customer is not modelled. | P(buys(c1,h2)) = [no:0.47,yes:0.53]
P(buys(c1,h2)|rich(c1)=no,cheap(h2)=no) = [no:0.744,yes:0.256] |
| realestate.blog | realestate encoding BLOG | |
| realestate.yap | realestate encoding for CLP(BN) | P(buys(B,gates,villa)|cheap(no,villa))=? |
| realestate.ibl |
Implementation of the realestate Domain.
Remark: The implementation has several Problems.
First it does not keep track of the result of
queries. this means that for example wants(customer,facility)
will lead in two different calls in two different results.
Second it is not possible to pose any usefull query
as almost all give "Fatal error: exception Stack_overflow".
e.g.
fac1=facility()
price(Cons(fac1,Nil))
works, but when calling this function via
house(Cons(fac1,Nil))
it crashes
| |
realestate.psm
realestate_task1_extra.psm | PRISM encoding | |
realestate.rbn
realestate.rst | realestate encoding PRIMULA | P(expensive(house2)) = 0.1865
P(expensive(house2) | has(house2,swimmingPool)) = 0.4024 M
Sampling-based inference implemented in Primula. |
- Weather Markov Model (Leuven Exercise)
- Weather Hidden Markov Model (Leuven Exercise)
- Weather Hidden Markov Model Where Umbrella Influences the Weather (Leuven Exercise)
- Weather Hidden Markov Model With Several Guards (Leuven Exercise)
- Weather Hidden Markov Model With Several Guards (Leuven Exercise)
|