I get the following statistics in Z3.
(:added-eqs 24529
:binary-propagations 43837
:bv-bit2core 7115
:bv-conflicts 156
:bv-diseqs 10395
:bv-dynamic-diseqs 10028
:bv->core-eq 10401
:conflicts 409
:decisions 4840
:del-clause 84926
:final-checks 2
:max-generation 4
:memory 5.69
:minimized-lits 467
:mk-clause 88358
:propagations 90195
:quant-instantiations 3388
:restarts 3
:time 0.83)
I'd like to know what the metrics used for each result row.
Can you help me?
Disclaimer: I have the feeling that interpreting the statistics the right way is quite an art, and that the Z3 developers are probably the only ones who really know how to do that. Anyway, here is what I know ... or believe:
quant-instantiations indicates the number of instantiated quantifiers. The fewer instantiations the better, but you of course don't want to make your patterns/triggers too strict because Z3 then won't be able to prove anything.
conflicts indicate assignments that happen in the theory subsolvers and that did not make the formula true. If the formula can be satisfied and the number of conflicts is high, it basically means that the prover tried lots of assignments that did not satisfy the formula, i.e., that the prover did not manage to explore the search space in the direction of the goal.
Related questions:
Z3 statistics: what does time measure?
Z3 real arithmetic and statistics
Which statistics indicate an efficient run of Z3?
Related
I have been working with the optimizer in Z3PY, and only using Z3 ints and (x < y)-like constraints in my project. It has worked really well. I have been using up to 26 variables (Z3 ints), and it takes the solver about 5 seconds to find a solution and I have maybe 100 soft constraints, at least. But now I tried with 49 variables, and it does not solve it at all (I shut it down after 1 hour).
So I made a little experiment to find out what was slowing it down, is it the amount of variables or the amount of soft constraints? It seems like the bottle neck is the amount of variables.
I created 26 Z3-ints. Then I added as hard constraints, that it should not be lower than 1 or more than 26. Also, all numbers must be unique. No other constraints was added at all.
In other words, the solution that the solver will find is a simple order [1,2,3,4,5....up to 26]. Ordered in a way that the solver finds out.
I mean this is a simple thing, there are really no constraints except those I mentioned. And the solver solves this in 0.4 seconds or something like that, fast and sufficient. Which is expected. But if I increase the amount of variables to 49 (and of course the constraints now are that it should not be lower than 1 or more than 49), it takes the solver about 1 minute to solve. That seems really slow for such a simple task? Should it be like this, anybody knows? The time complexity is really extremely increased?
(I know that I can use Solver() instead of Optimizer() for this particular experiment, and it will be solved within a second, but in reality I need it to be done with Optimizer since I have a lot of soft constraints to work with.)
EDIT: Adding some code for my example.
I declare an array with Z3 ints that I call "reqs".
The array is consisting of 26 variables in one example and 49 in the other example I am talking about.
solver = Optimize()
for i in (reqs):
solver.add(i >= 1)
for i in (reqs):
solver.add(i <= len(reqs))
d = Distinct(reqs)
solver.add(d)
res = solver.check()
print(res)
Each benchmark is unique, and it's impossible to come up with a good strategy that applies equally well in all cases. But the scenario you describe is simple enough to deal with. The performance problem comes from the fact that Distinct creates too many inequalities (quadratic in number) for the solver, and the optimizer is having a hard time dealing with them as you increase the number of variables.
As a rule of thumb, you should avoid using Distinct if you can. For this particular case, it'd suffice to impose a strict ordering on the variables. (Of course, this may not always be possible depending on your other constraints, but it seems what you're describing can benefit from this trick.) So, I'd code it like this:
from z3 import *
reqs = [Int('i_%d' % i) for i in range(50)]
solver = Optimize()
for i in reqs:
solver.add(i >= 1, i <= len(reqs))
for i, j in zip(reqs, reqs[1:]):
solver.add(i < j)
res = solver.check()
print(res)
print(solver.model())
When I run this, I get:
$ time python a.py
sat
[i_39 = 40,
i_3 = 4,
...
i_0 = 1,
i_2 = 3]
python a.py 0.27s user 0.09s system 98% cpu 0.365 total
which is pretty snippy. Hopefully you can generalize this to your original problem.
I using Z3 C++ api to find a satisfiable formula that is minimal with respect to some boolean variables (let us call them b0,...,bn) being true.
I have a formula that includes boolean variables b0,...,bn and I want to find some satisfiable formula where I have the least number of b0,...,bn set to true.
I do this by initially finding a subset of b0,...,bn that can be assigned to true and satisfy my formula, and I incrementally ask the solver to find smaller subsets (i.e. where one of these boolean variables is flipped to false).
I find my local minimum when I cannot find a smaller subset, i.e. I get a unsat result from the Z3. At this point, I would like to access the last valid model.
Is that possible? Does Z3 modify the model when a call to "check" is unsat?
If so, how can I do this using the C++ api?
Many thanks in advance,
You can retrieve a model if the solver returns "sat". The model refers to the state of the solver, so if you add assertions, the state changes and models are no longer valid until you check satisfiability and it returns sat.
So you can retrieve a model every time the solver returns SAT, and then discharge all but the last model.
As Nikolaj mentioned, you need to keep track of models after each call that results in sat and return the last one when you get an unsat if you follow the strategy you outlined.
However, there might be another alternative that avoids repeated calls altogether. Instead of a satisfaction problem, you can cast your problem as an optimization one. You mentioned you have control variables b0, b1, .. bn such that you want to minimize the number of them getting set to true for a satisfying model. Create a metric that counts the number of ones in these variables. Something like:
metric = (if b0 then 1 else 0)
+ (if b1 then 1 else 0)
+ ...
+ (if bn then 1 else 0)
Then use Z3's optimization routines to minimize metric. I believe this will provide you with the solution you are looking for in one call only.
Some helpful references:
Here's the Z3 optimization tutorial: http://rise4fun.com/z3opt/tutorialcontent/guide.
This example, in particular, talks about soft-constraints, and might quite be applicable in your case as well: http://rise4fun.com/z3opt/tutorialcontent/guide#h25.
Here's the C++ API reference for the optimizer: http://z3prover.github.io/api/html/classz3_1_1optimize.html.
I start with a simple Maxima question, the answer to which may provide the answer to the actual problem I'm grappling with.
Related Simple Question:
How can I get maxima to calculate:
bfloat((1+%i)^0.3);
Might there be an option variable that can be set so that this evaluates to a complex number?
Actual Question:
In evaluating approximations for numerical time integration for finite element methods, for this purpose I'm using spectral analysis, which requires the calculation of the eigenvalues of a 4 x 4 matrix. This matrix "cav" is also calculated within maxima, using some of the algebra capabilities of maxima, but sustituting numerical values, so that matrix is entirely numerical, i.e. containing no variables. I've calculated the eigenvalues with Mathematica and it returns 4 real eigenvalues. However Maxima calculates horrenduously complicated expressions for this case, which apparently it does not "know" how to simplify, even numerically as "bigfloat". Perhaps this problem arises because Maxima first approximates the matrix "cac" by rational numbers (i.e. fractions) and then tries to solve the problem fully exactly, instead of simply using numerical "bigfloat" computations throughout. Is there I way I can change this?
Note that if you only change the input value of gzv to say 0.5 it works fine, and returns numerical values of complex eigenvalues.
I include the code below. Note that all of the code up until "cav:subst(vs,ca)$" is just for the definition of the matrix cav and seems to work fine. It is in the few statements thereafter that it fails to calculate numerical values for the eigenvalues.
v1:v0+ (1-gg)*a0+gg*a1$
d1:d0+v0+(1/2-gb)*a0+gb*a1$
obf:a1+(1+ga)*(w^2*d1 + 2*gz*w*(d1-d0)) -
ga *(w^2*d0 + 2*gz*w*(d0-g0))$
obf:expand(obf)$
cd:subst([a1=1,d0=0,v0=0,a0=0,g0=0],obf)$
fd:subst([a1=0,d0=1,v0=0,a0=0,g0=0],obf)$
fv:subst([a1=0,d0=0,v0=1,a0=0,g0=0],obf)$
fa:subst([a1=0,d0=0,v0=0,a0=1,g0=0],obf)$
fg:subst([a1=0,d0=0,v0=0,a0=0,g0=1],obf)$
f:[fd,fv,fa,fg]$
cad1:expand(cd*[1,1,1/2-gb,0] - gb*f)$
cad2:expand(cd*[0,1,1-gg,0] - gg*f)$
cad3:expand(-f)$
cad4:[cd,0,0,0]$
cad:matrix(cad1,cad2,cad3,cad4)$
gav:-0.05$
ggv:1/2-gav$
gbv:(ggv+1/2)^2/4$
gzv:1.1$
dt:0.01$
wv:bfloat(dt*2*%pi)$
vs:[ga=gav,gg=ggv,gb=gbv,gz=gzv,w=wv]$
cav:subst(vs,ca)$
cav:bfloat(cav)$
evam:eigenvalues(cav)$
evam:bfloat(evam)$
eva:evam[1]$
The main problem here is that Maxima tries pretty hard to make computations exact, and it's hard to tell it to ease up and allow inexact results.
Is there a mistake in the code you posted above? You have cav:subst(vs,ca) but ca is not defined. Is that supposed to be cav:subst(vs,cad) ?
For the short problem, usually rectform can simplify complex expressions to something more usable:
(%i58) rectform (bfloat((1+%i)^0.3));
`rat' replaced 1.0B0 by 1/1 = 1.0B0
(%o58) 2.59023849130283b-1 %i + 1.078911979230303b0
About the long problem, if fixed-precision (i.e. ordinary floats, not bigfloats) is acceptable to you, then you can use the LAPACK function dgeev to compute eigenvalues and/or eigenvectors.
(%i51) load (lapack);
<bunch of messages here>
(%o51) /usr/share/maxima/5.39.0/share/lapack/lapack.mac
(%i52) dgeev (cav);
(%o52) [[- 0.02759949957202372, 0.06804641655485913, 0.997993508502892, 0.928429191717788], false, false]
If you really need variable precision, I don't know what to try. In principle it's possible to rework the LAPACK code to work with variable-precision floats, but that's a substantial task and I'm not sure about the details.
I want to port a 32 by 32 bit unsigned multiplication on a 24-bit dsp (it's a Linear Congruential Generator, so I'm not allowed to truncate, also I don't want to replace yet the current LCG with a 24 bit one). The available data types are 24 and 48 bit ints.
Only the last 32 LSB are needed. Do you know any hacks to implement this in fewer multiplies, masks and shifts than the usual way?
The line looks like this:
//val is an int(32 bit)
val = (1664525 * val) + 1013904223;
An outline would be (in my current compiler style):
static uint48_t val = SEED;
...
val = 0xFFFFFFFFUL & ((1664525UL * val) + 1013904223UL);
and hopefully the compiler will recognise:
it can use a multiply and accumulate command
it only needs a reduced multiply algorithim due to the "high word" of the constant being zero
the AND could be effected by resetting the upper bits or multiplying a constant and restoring
...other stuff depends on your {mystery dsp} target
Note
if you scale up the coefficients by 2^16, you can get truncation for free, but due to lack of info
you will have to explore/decide if it is better overall.
(This is more an elaboration why two multiplications 24×24→n, 31<n are enough for 32×32→min(n, 40).)
The question discloses amazingly little about the capabilities to build a method
32×21→32 in fewer [24×24] multiplies, masks and shifts than the usual way on:
24 and 48 bit ints & DSP (I read high throughput, non-high latency 24×24→48).
As far as there indeed is a 24×24→48 multiply (or even 24×24+56→56 MAC) and one factor is less than 24 bits, the question is pointless, a second multiply being the compelling solution.
The usual composition of a 24<n<48×24<m<48→24<p multiply from 24×24→48 uses three of the latter; a compiler should know as well as a coder that "the fourth multiply" would yield bits with a significance/position exceeding the combined lengths of the lower parts of the factors.
So, is it possible to generate "the long product" using just a second 24×24→48?
Let the (bytes of the) factors be w_xyz and W_XYZ, respectively; the underscores suggesting "the Ws" being the lower significance bits in the higher significance words/ints if interpreted as 24bit ints. The first 24×24→48 gives the sum of
zX
yXzY
xXyYzZ
xYyZ
xZ, what is needed (fat) is
wZ +
zW.
This can be computed using one combined multiplication of
((w<<16)|(z & 0xff)) × ((W<<16)|(Z & 0xff)). (Never mind the 17th bit of wZ+zW "running" into wW.)
(In the first revision of this answer, I foolishly produced wZ and zW separately - their sum is wanted in the end, anyway.)
(Annoyingly, this is about all you can do for 24×24→24 as a base operation too - beyond this "combining multiplication", you need four instead of one.)
Another angle to explore is choosing a different PRNG.
It may have to be >24 bits (tell!).
On a 24 bit machine, XorShift* (or even XorShift+) 48/32 seems worth a look.
I was wondering if anyone knew any intuitive crossover and mutation operators for paths within a graph? Thanks!
Question is a bit old, but the problem doesn't seem to be outdated or solved, so I think my research still might be helpful for someone.
As far as mutation and crossover is quite trivial in the TSP problem, where every mutation is valid (that is because chromosome represents an order of visiting fixed nodes - swapping order then always can create a valid result), in case of Shortest Path or Optimal Path, where the chromosome is a exact route representation, this doesn't apply and isn't that obvious. So here is how I approach problem of solving Optimal Path using GA.
For crossover, there are few options:
For routes that have at least one common point (besides start and end node) - find all common points and swap subroutes in the place of crossing
Parent 1: 51 33 41 7 12 91 60
Parent 2: 51 9 33 25 12 43 15 60
Potential crossing point are 33 and 12. We can get following children: 51 9 33 41 7 12 43 15 60 and 51 33 25 12 91 60 that are the result of crossing using both of these crossing points.
When two routes don't have common point, select randomly two points from each parent and connect them (you can use for that either random traversal, backtracking or heuristic search like A* or beam search). Now this path may be treated as crossover path. For better understanding, see below picture of two crossover methods:
see http://i.imgur.com/0gDTNAq.png
Black and gray paths are parents, pink and orange paths are
children, green point is a crossover place, and red points are start
and end nodes. First graph shows first type of crossover, second graph is example of another one.
For mutation, there are also few options. Generally, dummy mutation like swapping order of nodes or adding random node is really ineffective for graphs with average density. So here are the approaches that guarantee valid mutations:
Take randomly two points from path and replace them with a random path between those two nodes.
Chromosome: 51 33 41 7 12 91 60 , random points: 33 and 12, random/shortest path between then: 33 29 71 12, mutated chromosome: 51 33 29 71 12 91 60
Find random point from path, remove it and connect its neighbours (really very similar to the first one)
Find random point from path and find random path to its neighbour
Try subtraversing the path from some randomly chosen point, until reaching any point on the initial route (slight modification of the first method).
see http://i.imgur.com/19mWPes.png
Each graph corresponds to each mutation method in appropriate order. In last example, the orange path is the one that would replace original path between mutation points (green nodes).
Note: this methods obviously may have performance drawback in the case, when finding alternative subroute (using a random or heuristic method) will stuck at some place or find very long and useless subpath, so consider bounding the time of mutation execution or trials number.
For my case, which is finding an optimal path in terms of maximizing sum of vertices weights while keeping sum of nodes weight less than given bound, those methods are quite effective and give a good result. Should you have any question, feel free to ask. Also, sorry for my MS Paint skills ;)
Update
One big hint: I basically used this approach in my implementation, but there was one big drawback of using random path generating. I decided to switch to semi-random route generation using shortest path traversing randomly picked point(s) - it is much more efficent (but obviously may not be applicable for all problems).
Emm.. That is very difficult question, people write dissertations for that and still there is no right answer to that.
The general rule is "it all depends on your domain".
There are some generic GA libraries that will do some work for you, but for the best results it is recommended to implement your GA operations yourself, specifically for your domain.
You might have more luck with answers on Theoretical CS, but you need to expand your question more and add more details about your task and domain.
Update:
So you have a graph. In GA terms, a path through the graph represents an individual, nodes in the path would be chromosomes.
In that case I would say a mutation can be represented as deviation of the path somewhere from the original - one of the nodes is moved somewhere, and the path is adjusted so the start and end values in the path are remaining the same.
Mutation can lead to invalid individuals. And in that case you need to make a decision: allow invalid ones and hope that they will converge to some unexplored solution. Or kill them on the spot. When I was working with GA, I did allow invalid solution, adding "Unfitness" value along with fitness. Some researchers suggest this can help with broad exploring of the solution space.
Crossover can only happen to the paths that are crossing each other: on the point of the crossing, swap the remains of the path with the parents.
Bear in mind that there are various ways for crossover: individuals can be crossed-over in multiple points or just in one. In the case with graphs you can have multiple crossing points, and that can naturally lead to the multiple children graphs.
As I said before, there is no right or wrong way of doing this, but you will find out the best way only by experimenting on it.