⚠️ Note: This is not the commercial version of Hypertrie, but a research prototype that focuses on aspects presented in papers. For a reliable, SPARQL feature-complete, and well-tested edition, see the commercial edition at https://github.com/tentris/tentris.

Hypertrie

Where is it used?

• the backend of the commercial and research-prototype Tentris triple stores/RDF graph databases
• a sparse tensor representation that supports slicing by any dimension and einstein summation
• a monolithic index that supports worst-case optimal joins (WCOJ) by providing all collation orders in a single redundancy eliminating datastructure.

What is it?

Technically, a hypertrie stores $d$-tuples where $d$ is also called dimension (tensor) or depth (trie, index).

It allows incremental slicing (tensor) by any dimension or select and project by any predicate (relational algebra). These properties are important to support worst-case optimal joins (WCOJ) efficiently.

Asymptotic Guarantees

A depth $d$ the hypertrie encoding a set of $z$ tuples requires at most $\mathcal O (z\cdot 2^{d-1}\cdot d)$ space. The runtime complexity of applying (inserting or deleting) a changeset set $\Delta$ of $d$-tuples to a depth $d$ hypertrie is bound by the space complexity $\mathcal O (|\Delta| \cdot 2^{d-1}\cdot d)$ of a surrogate hypertrie that encodes the change set changeset $\Delta$.

build

prerequisites

install conan, cmake and a C++20 compiler.

and create a conan profile

conan profile new --detect default
conan profile update settings.compiler.libcxx=libstdc++11 default

You'll need some packages from DICE group's conan artifactory. Add it with:

conan remote add dice-group https://conan.dice-research.org/artifactory/api/conan/tentris

build

mkdir build
cd build
conan install .. --build=missing
cmake ..

running tests

To enable test, set DBUILD_TESTING in cmake:

cmake -DBUILD_TESTING=ON ..
make -j tests
tests/tests

Some tests are using pytorch which is not provided with the code. Those tests are disabled by default. To enable them, provide the path to the pytorch library via cmake variable hypertrie_LIBTORCH_PATH. Prebuild binaries may be downloaded via https://pytorch.org/get-started/locally/ (works at least with Stable|Linux|LibTorch|C++|None).

cmake -DDBUILD_TESTING=ON -DLIBTORCH_PATH=/path/to/libtorch ..