Add GFD mining #465
Add GFD mining #465
Conversation
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
AntonChern
force-pushed
the
gfd
branch
2 times, most recently
from
c3eeecf to
a59e3cc
Compare
| : query_(query_), iso_(iso_), res_(res_) {} | ||
|
|
||
| template <typename CorrespondenceMap1To2, typename CorrespondenceMap2To1> | ||
| bool operator()(CorrespondenceMap1To2 f, CorrespondenceMap2To1) const { |
There was a problem hiding this comment.
Second parameter is not used
There was a problem hiding this comment.
Boost Graph Library requires such syntax from callback function. Unfortunately, if I change the function signature, the code will not compile
|
Add PR description |
|
You are doing great! Please don't forget to notify when you are done with changes requested. Also please mark conversations if they are resolved |
AntonChern
force-pushed
the
gfd
branch
2 times, most recently
from
60aeadc to
9ce90a3
Compare
| Token name_token = Token(i, name.first); | ||
| for (auto& value : name.second) { | ||
| Token value_token = Token(-1, value); | ||
| result.push_back(Literal(name_token, value_token)); |
There was a problem hiding this comment.
result.reserve(name.second.size() * attrs_info.at(label).size())
There was a problem hiding this comment.
Unfortunately, name.second can have different number of values, depending on name.first. The reserve may not work correctly.
AntonChern
force-pushed
the
gfd
branch
2 times, most recently
from
11d6a2e to
13b76ae
Compare
AntonChern
force-pushed
the
gfd
branch
2 times, most recently
from
b5fd5d6 to
ea596c9
Compare
AntonChern
force-pushed
the
gfd
branch
2 times, most recently
from
9e67961 to
8cb290f
Compare
|
LGTM |
| for (std::size_t i = 0; i < patterns.size(); ++i) { | ||
| auto pattern = patterns[i]; |
There was a problem hiding this comment.
I don't know if we have std::views::enumerate support, but if we do, than it's better to use it in cases where we need both an index and an element:
| for (std::size_t i = 0; i < patterns.size(); ++i) { | |
| auto pattern = patterns[i]; | |
| for (auto [i, pattern] : std::views::enumerate(patterns)) { |
| for (std::size_t i = 0; i < patterns.size(); ++i) { | ||
| graph_t pattern = patterns.at(i); |
There was a problem hiding this comment.
Same here if we have std::views::enumerate
There was a problem hiding this comment.
As far as I understand, the project only supports C++20
There was a problem hiding this comment.
I didn't look at almost anything, the logic part was hopefully checked by other reviewers. I only looked at the tests.
I described the most important comment in the test file itself. But I would also like to draw attention to two things:
- Why do we have access to types specific to this primitive (
Gfd,Literal,Token) in thealgosnamespace? Usually themodelnamespace is used for this.LiteralandTokenmust definitely not be in this namespace, andGfdis still acceptable, but also not very good. - Some tests run for quite a long time in debug mode:
https://github.com/Desbordante/desbordante-core/actions/runs/14045967659/job/39326756580?pr=465
Run core-tests on ubuntu-latest with llvm-clang, Debug
270/719 Test #273: GfdMiningTest.TestMovies ........................................................................................................................................................................................................................................................................................................................... Passed 13.92 sec
Start 274: GfdMiningTest.TestSymbols
271/719 Test #274: GfdMiningTest.TestSymbols .......................................................................................................................................................................................................................................................................................................................... Passed 129.41 sec
Start 275: GfdMiningTest.TestShapes
272/719 Test #275: GfdMiningTest.TestShapes ........................................................................................................................................................................................................................................................................................................................... Passed 61.20 sec
Is this normal? Are there known limits that are acceptable? It seems that this is quite long. Such tests are usually marked with an additional prefix (HeavyDatasets), so as not to run too expensive tests every time.
Add class for GFD miner
Add minimal GFD and GFD with multiple conclusion tests
Added two examples with searching for dependencies in small graphs.
This PR implements an algorithm for mining graph functional dependencies based on article "Discovering Graph Functional Dependencies" by Fan Wenfei, Hu Chunming, Liu Xueli, and Lu Ping. Algorithm, given an input graph, returns a set of dependencies satisfied on this graph. The algorithm also has two configurable parameters:
kis the maximum number of vertices in the pattern of the mined dependency andsigmais its minimum frequency.In addition, the PR implements the ability to run the algorithm in Python, and also contains examples of its use.