LaTiS: Enabling Interoperability via a Universal Functional Data Model

Abstract: 

LaTiS is a software framework for data access, processing, and output. The modular architecture supports reusable and custom Readers to read a dataset from its native source, Operations to manipulate the dataset, and Writers to output the dataset in the desired form. Datasets can be read from diverse sources, combined in various ways to derive new datasets, and written to any number of formats. LaTiS can enable simple access to a single data file or it can be used to orchestrate an entire data processing workflow.

 

The core feature of LaTiS that enables these capabilities is its Functional Data Model. This data model extends the Relational Data Model to add the concept of Functional Relationships between independent and dependent variables which are prevalent in scientific data. This model provides a mathematical foundation for describing any dataset in terms of only three variable types:

Scalar: A single Variable.
Tuple: A group of Variables.
Function: A mapping from one Variable (domain) to another (range).
 

Since Variables can be any one of these three types, they can be composed in arbitrarily complex ways to represent the underlying nature of any dataset. LaTiS does not simply provide an alternate data model, it can be used to represent the fundamental mathematical structure of any other data model. In other words, any data model can be mapped to the lower level LaTiS Data Model. The Variable types can then be extended to add semantics that are specific to a particular scientific discipline (e.g. Temperature as a specific type of Scalar). The resulting datasets can still be used by those outside that discipline, albeit with a loss of the extra semantics, because the model representation can always be reduced to the basic three variable types. In this way, LaTiS provides a new level of interoperability between disparate datasets.

 

The Open Source Scala implementation of the LaTiS Data Model (https://github.com/dlindhol/LaTiS) is designed to serve as a Domain Specific Language (DSL) for scientific data analysis. The version 2.x series is in part a testbed for the evolution of the LaTiS DSL. Version 3 of LaTiS will provide a full featured interactive data analysis tool using the Scala REPL (command-line interface) to build on the ability of LaTiS to represent scientific datasets with higher level Functional semantics (e.g. time series of grids) as opposed to the more commonly used multi-dimensional array constructs. Combining the Functional semantics of the data model with the Functional Programming constructs of Scala promises to be quite powerful.

 

Perhaps the current "killer app" for the LaTiS framework is its web service interface to the underlying LaTiS DSL. This RESTful API implements the standardized OPeNDAP (DAP2) data request and reply specification. Data providers can easily install a LaTiS Server and expose the datasets they wish to serve by creating TSML dataset descriptors. The service layer will accept queries that include selection constraints (e.g. time > 2014-05-01), as opposed to requiring array index notation, and function calls (e.g. format_time(yyyyDDD)) that can subset, aggregate, and perform other operations on the server before writing the resulting dataset to the client. 

 

This poster will provide an overview of the LaTiS Data Model and the Open Source Scala implementation and service architecture with the following goals:

  • Enable data providers to run their own LaTiS Server to provide broader access to their data
  • Enable the user community to make new discoveries via improved data interoperability
  • Build a LaTiS developer community to strengthen the core libraries and to increase the number of sharable components

 

 

Attachments for download: 
Creative Common License: 
Creative Commons Attribution 3.0 License