jena-datatensor

jena-datatensor is an extension for Apache Jena that introduces 2 new datatypes, 36 SPARQL functions, and 6 new aggregators to enhance the processing of data tensors within RDF models.

exemplary dataset with queries contains a sligthly modified subset of CS publications knowledge graph, which was a part of bachelor thesis and queries to demonstrate the usage of the implemented functions and aggregators. This dataset is under fair use license and is provided for educational purposes only.

Features

• New Datatypes:
  • dt:NumericDataTensor – Designed for storing tensors containing numeric values.
  • dt:BooleanDataTensor – Intended for storing tensors containing boolean values.
• 36 New SPARQL Functions:
  • Tensor manipulations (addition, multiplication, reshaping, etc.)
  • Algebraic computations
• 6 New Aggregators:
  • Generalized aggregation functions for numerical tensors
  • Sum, average, variance, and standard deviation computations

Installation

1. Clone the repository:

   git clone https://github.com/cinekele/jena-datatensor.git
   cd jena-datatensor/jena-datatensor

2. Build the project:

   mvn install

3. Add the generated JAR to your Jena classpath or include it in your project dependencies.

Implemented Functions

The table below lists the implemented SPARQL functions, including their input arguments, return values, and associated IRI (with the dtf prefix set to http://example.org/date-tensor/functions#).

Transforming Functions

IRI	Function	Input Arguments	Return Value
dtf:cos	Applies cosine function	One numerical tensor	Tensor with same shape
dtf:exp	Applies exponential function	One numerical tensor	Tensor with same shape
dtf:log	Applies natural logarithm	One numerical tensor	Tensor with same shape
dtf:logp	Logarithm with base p	Number (p), numerical tensor	Tensor with same shape
dtf:poly	Raises elements to power	Number, numerical tensor	Tensor with same shape
dtf:scale	Scales tensor elements	Number, numerical tensor	Tensor with same shape
dtf:sin	Applies sine function	One numerical tensor	Tensor with same shape
dtf:abs	Applies absolute value	One numerical tensor	Tensor with same shape

Operators

IRI	Function	Input Arguments	Return Value
dtf:not	Logical NOT operation	One boolean tensor	Boolean tensor
dtf:add	Element-wise addition	Two numerical tensors	Numerical tensor (broadcasted)
dtf:subtract	Element-wise subtraction	Two numerical tensors	Numerical tensor (broadcasted)
dtf:multiply	Element-wise multiplication	Two numerical tensors	Numerical tensor (broadcasted)
dtf:divide	Element-wise division	Two numerical tensors	Numerical tensor (broadcasted)
dtf:eq	Element-wise equality check	Two tensors	Boolean tensor (broadcasted)
dtf:neq	Element-wise inequality check	Two tensors	Boolean tensor (broadcasted)
dtf:and	Logical AND operation	Two boolean tensors	Boolean tensor (broadcasted)
dtf:or	Logical OR operation	Two boolean tensors	Boolean tensor (broadcasted)
dtf:gt	Element-wise greater than comparison	Two numerical tensors	Boolean tensor (broadcasted)
dtf:lt	Element-wise less than comparison	Two numerical tensors	Boolean tensor (broadcasted)

Indexing Functions

IRI	Function	Input Arguments	Return Value
dtf:getSubDT	Extracts sub-tensor	One numerical tensor, second tensor (numerical/boolean)	Sub-tensor

Concatenating Functions

IRI	Function	Input Arguments	Return Value
dtf:concat	Concatenates tensors	Two tensors, optional dimension	Combined tensor
dtf:hstack	Horizontally stacks tensors	Two tensors	Horizontally combined tensor
dtf:vstack	Vertically stacks tensors	Two tensors	Vertically combined tensor

Reduction Functions

IRI	Function	Input Arguments	Return Value
dtf:all	Checks if all elements are true	Boolean tensor	Boolean scalar
dtf:any	Checks if any element is true	Boolean tensor	Boolean scalar
dtf:avg	Computes average	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:sum	Computes sum	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:max	Computes maximum	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:median	Computes median	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:min	Computes minimum	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:std	Computes standard deviation	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:var	Computes variance	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:norm1	Computes L1 norm	Axis (number), numerical tensor	Reduced tensor (or scalar)
dtf:norm2	Computes L2 norm	Axis (number), numerical tensor	Reduced tensor (or scalar)

Similarity Functions

IRI	Function	Input Arguments	Return Value
dtf:cosineSimilarity	Computes cosine similarity	Two numerical tensors	Numeric scalar
dtf:euclideanDistance	Computes Euclidean distance	Two numerical tensors	Numeric scalar

Implemented Aggregators

The table below lists the implemented aggregation functions, which operate on NumericDataTensor inputs and return NumericDataTensor results. These aggregators generalize standard aggregation functions for numerical tensors and do not support the DISTINCT keyword. The dta prefix set to http://example.org/datatensor/functions/aggregates#)

IRI (with Turtle prefix)	Description
dta:sum	Sums grouped tensors
dta:avg	Calculates an average tensor
dta:var	Computes the variance tensor
dta:std	Computes the standard deviation tensor