DOCSLIB.ORG

A Relational Matrix Algebra and Its Implementation in a Column Store

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Relational Matrix Algebra and Its Implementation in a Column Store Zurich Open Repository and Archive University of Zurich Main Library Strickhofstrasse 39 CH-8057 Zurich www.zora.uzh.ch Year: 2020 A Relational Matrix Algebra and its Implementation in a Column Store Dolmatova, Oksana ; Augsten, Nikolaus ; Böhlen, Michael Hanspeter Abstract: Analytical queries often require a mixture of relational and linear algebra operations applied to the same data. This poses a challenge to analytic systems that must bridge the gap between relations and matrices. Previous work has mainly strived to fix the problem at the implementation level. This paper proposes a principled solution at the logical level. We introduce the relational matrix algebra (RMA), which seamlessly integrates linear algebra operations into the relational model and eliminates the dichotomy between matrices and relations. RMA is closed: All our relational matrix operations are performed on relations and result in relations; no additional data structure is required. Our implementa- tion in MonetDB shows the feasibility of our approach, and empirical evaluations suggest that in-database analytics performs well for mixed workloads. DOI: https://doi.org/10.1145/3318464.3389747 Posted at the Zurich Open Repository and Archive, University of Zurich ZORA URL: https://doi.org/10.5167/uzh-200904 Conference or Workshop Item Published Version Originally published at: Dolmatova, Oksana; Augsten, Nikolaus; Böhlen, Michael Hanspeter (2020). A Relational Matrix Algebra and its Implementation in a Column Store. In: ACM SIGMOD International Conference on Management of Data, Portland, OR, USA, 14 June 2020 - 19 June 2020, 2573-2587. DOI: https://doi.org/10.1145/3318464.3389747 A Relational Matrix Algebra and its Implementation in a Column Store Oksana Dolmatova Nikolaus Augsten Michael H. Böhlen University of Zürich University of Salzburg University of Zürich Zürich, Switzerland Salzburg, Austria Zürich, Switzerland [email protected] [email protected] [email protected] ABSTRACT Dealing with mixed workloads is challenging since the Analytical queries often require a mixture of relational and gap between relations and matrices must be bridged. Current linear algebra operations applied to the same data. This poses relational systems are poorly equipped for this task. Previ- a challenge to analytic systems that must bridge the gap be- ous attempts to deal with mixed workloads have focused tween relations and matrices. Previous work has mainly on the implementation level, for example, by introducing strived to fix the problem at the implementation level. This ordered data types; by storing matrices in special relations or paper proposes a principled solution at the logical level. We key-value structures; or by splitting queries into their rela- introduce the relational matrix algebra (RMA), which seam- tional and matrix parts. This paper resolves the gap between lessly integrates linear algebra operations into the relational relations and matrices. model and eliminates the dichotomy between matrices and We propose a principled solution for mixed workloads relations. RMA is closed: All our relational matrix operations and introduce the relational matrix algebra (RMA) to support are performed on relations and result in relations; no ad- complex data analysis within the relational model. The goal ditional data structure is required. Our implementation in is to (1) solve the integration of relations and linear algebra at MonetDB shows the feasibility of our approach, and empiri- the logical level, (2) so achieve independence from the imple- cal evaluations suggest that in-database analytics performs mentation at the physical level, and (3) prove the feasibility well for mixed workloads. of our model by extending an existing system. We are the first to achieve these goals: Other works focus on facilitating ACM Reference Format: the transition between the relational and the linear algebra Oksana Dolmatova, Nikolaus Augsten, and Michael H. Böhlen. 2020. model. We eliminate the dichotomy between matrices and A Relational Matrix Algebra and its Implementation in a Column Store. In Proc of the 2020 ACM SIGMOD International Conference on relations by seamlessly integrating linear algebra into the Management of Data, June 14-19, 2020, Portland, OR, USA. ACM, New relational model. Our implementation of RMA in MonetDB York, NY, USA, 15 pages. https://doi.org/10.1145/3318464.3389747 shows the feasibility of our approach. We define linear operations over relations and system- 1 INTRODUCTION atically process and maintain non-numerical information. We show that the relational model is well-suited for com- Many data that are stored in relational databases include plex data analysis if ordering and contextual information are numerical parts that must be analyzed, for example, sensor dealt with properly. RMA is purely based on relations and data from industrial plants, scientific observations, or point does not introduce any ordered data structures. Instead, the of sales data. The analysis of these data, which are not purely relevant row order for matrix operations is computed from numerical but also include important non-numerical val- contextual information in the argument relations. All rela- ues, demand mixed queries that apply relational and linear tional matrix operations return relations with origins. Origins algebra operations on the same data. are constructed from the contextual information (attribute Permission to make digital or hard copies of all or part of this work for names and non-numerical values) of the input relations and personal or classroom use is granted without fee provided that copies are not uniquely identify and describe each cell in the result relation. made or distributed for profit or commercial advantage and that copies bear We extend the syntax of SQL to support relational matrix this notice and the full citation on the first page. Copyrights for components operations. As an example, consider a relation 푟푎푡푖푛푔 with of this work owned by others than ACM must be honored. Abstracting with schema (푈푠푒푟, 퐵푎푙푡표, 퐻푒푎푡, 푁푒푡), that stores users and their credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request ratings for the three films ("Balto", "Heat", and "Net", one permissions from [email protected]. column per film). The SQL query SIGMOD’20, June 14-19, 2020, Portland, OR, USA SELECT * FROM INV( rating BY User ); © 2020 Association for Computing Machinery. ACM ISBN 978-1-4503-6735-6/20/06...$15.00 orders the relation by users and computes the inversion of https://doi.org/10.1145/3318464.3389747 the matrix formed by the values of the ordered numerical columns. The result is a relation with the same schema: The The set of operations is limited and does not include opera- values of attribute 푈푠푒푟 are preserved, and the values of the tions whose results depend on the row order. For instance, remaining three attributes are provided by matrix inversion there are no SQL solutions for inversion or determinant com- (see Section 5 for details). The origin of a numerical result putation. Complex operations must be programmed as UDFs. value is given by the user name in its table row and the Ordonez et al. [26] suggest UDFs for linear regression with attribute name of its column. a matrix-like result type. The UTL_NLA package [25] for At the system level, we have integrated our solution into Oracle DBMS offers linear algebra operations defined over MonetDB. Specifically, we extended the kernel with rela- UTL_NLA_ARRAY. UDFs provide a technical interface but tional matrix operations implemented over binary associa- do not define matrix operations over relations. No systematic tion tables (BATs). The physical implementation of matrix approach to maintain contextual information is provided. operations is flexible and may be transparently delegated Luo et al. [20] extend SimSQL [8], a Hadoop-based rela- to specialized libraries that leverage the underlying hard- tional system, with linear algebra functionality. RasDaMan ware (e.g., MKL [2] for CPUs or cuBLAS [1] for GPUs). The [6, 22] manages and processes image raster data. Both sys- new functionality is introduced without changing the main tems introduce matrices as ordered numeric-only attribute data structures and the processing pipeline of MonetDB, and types. Although relations and matrices coexist, operations without affecting existing functionality. are defined over different objects. Linear operations arenot Our technical contributions are as follows: defined over unordered objects and they do not support con- • We propose the relational matrix algebra (RMA), which textual information for individual cells of a matrix. extends the relational model with matrix operations. SciQL [33, 34] extends MonetDB [23] with a new data type, This is the first approach to show that the relational ARRAY, as a first-class object. An array is stored as an object model is sufficient to support matrix operations. The on its own. Arrays have a fixed schema: The last attribute new set of operations is closed: All relational matrix stores the data values of a matrix, all other attributes are operations are performed on relations and result in dimension attributes and must be numeric. Arrays come relations, and no additional data structure is required. with a limited set of operations, such as addition, filtering, • We show that matrix operations are shape restricted, and aggregation, and they must be converted
Load more

Recommended publications
  • Informal Data Transformation Considered Harmful
    Informal Data Transformation Considered Harmful Eric Daimler, Ryan Wisnesky Conexus AI out the enterprise, so that data and programs that depend on that data need not constantly be re-validated for ev- ery particular use. Computer scientists have been develop- ing techniques for preserving data integrity during trans- formation since the 1970s (Doan, Halevy, and Ives 2012); however, we agree with the authors of (Breiner, Subrah- manian, and Jones 2018) and others that these techniques are insufficient for the practice of AI and modern IT sys- tems integration and we describe a modern mathematical approach based on category theory (Barr and Wells 1990; Awodey 2010), and the categorical query language CQL2, that is sufficient for today’s needs and also subsumes and unifies previous approaches. 1.1 Outline To help motivate our approach, we next briefly summarize an application of CQL to a data science project undertaken jointly with the Chemical Engineering department of Stan- ford University (Brown, Spivak, and Wisnesky 2019). Then, in Section 2 we review data integrity and in Section 3 we re- view category theory. Finally, we describe the mathematics of our approach in Section 4, and conclude in Section 5. We present no new results, instead citing a line of work summa- rized in (Schultz, Spivak, and Wisnesky 2017). Image used under a creative commons license; original 1.2 Motivating Case Study available at http://xkcd.com/1838/. In scientific practice, computer simulation is now a third pri- mary tool, alongside theory and experiment. Within
    [Show full text]
  • Automating the Capture of Data Transformations from Statistical Scripts in Data Documentation Jie Song George Alter H
    C2Metadata: Automating the Capture of Data Transformations from Statistical Scripts in Data Documentation Jie Song George Alter H. V. Jagadish University of Michigan University of Michigan University of Michigan Ann Arbor, Michigan Ann Arbor, Michigan Ann Arbor, Michigan [email protected] [email protected] [email protected] ABSTRACT CCS CONCEPTS Datasets are often derived by manipulating raw data with • Information systems → Data provenance; Extraction, statistical software packages. The derivation of a dataset transformation and loading. must be recorded in terms of both the raw input and the ma- nipulations applied to it. Statistics packages typically provide KEYWORDS limited help in documenting provenance for the resulting de- data transformation, data documentation, data provenance rived data. At best, the operations performed by the statistical ACM Reference Format: package are described in a script. Disparate representations Jie Song, George Alter, and H. V. Jagadish. 2019. C2Metadata: Au- make these scripts hard to understand for users. To address tomating the Capture of Data Transformations from Statistical these challenges, we created Continuous Capture of Meta- Scripts in Data Documentation. In 2019 International Conference data (C2Metadata), a system to capture data transformations on Management of Data (SIGMOD ’19), June 30-July 5, 2019, Am- in scripts for statistical packages and represent it as metadata sterdam, Netherlands. ACM, New York, NY, USA, 4 pages. https: in a standard format that is easy to understand. We do so by //doi.org/10.1145/3299869.3320241 devising a Structured Data Transformation Algebra (SDTA), which uses a small set of algebraic operators to express a 1 INTRODUCTION large fraction of data manipulation performed in practice.
    [Show full text]
  • What Is a Data Warehouse?
    What is a Data Warehouse? By Susan L. Miertschin “A data warehouse is a subject oriented, integrated, time variant, nonvolatile, collection of data in support of management's decision making process.” https: //www. bus iness.auc. dk/oe kostyr /file /What_ is_a_ Data_ Ware house.pdf 2 What is a Data Warehouse? “A copy of transaction data specifically structured for query and analysis” 3 “Data Warehousing is the coordination, architected, and periodic copying of data from various sources, both inside and outside the enterprise, into an environment optimized for analytical and informational processing” ‐ Alan Simon Data Warehousing for Dummies 4 Business Intelligence (BI) • “…implies thinking abstractly about the organization, reasoning about the business, organizing large quantities of information about the business environment.” p. 6 in Giovinazzo textbook • Purpose of BI is to define and execute a strategy 5 Strategic Thinking • Business strategist – Always looking forward to see how the company can meet the objectives reflected in the mission statement • Successful companies – Do more than just react to the day‐to‐day environment – Understand the past – Are able to predict and adapt to the future 6 Business Intelligence Loop Business Intelligence Figure 1‐1 p. 2 Giovinazzo • Encompasses entire loop shown Business Strategist • Data Storage + ETC = OLAP Data Mining Reports Data Warehouse Data Storage • Data WhWarehouse + Tools (yellow) = Extraction,Transformation, Cleaning DiiDecision Support CRM Accounting Finance HR System 7 The Data Warehouse Decision Support Systems Central Repository Metadata Dependent Data DtData Mar t EtExtrac tion DtData Log Administration Cleansing/Tranformation External Extraction Source Extraction Store Independent Data Mart Operational Environment Figure 1-2 p.
    [Show full text]
  • POLITECNICO DI TORINO Repository ISTITUZIONALE
    POLITECNICO DI TORINO Repository ISTITUZIONALE Rethinking Software Network Data Planes in the Era of Microservices Original Rethinking Software Network Data Planes in the Era of Microservices / Miano, Sebastiano. - (2020 Jul 13), pp. 1-175. Availability: This version is available at: 11583/2841176 since: 2020-07-22T19:49:25Z Publisher: Politecnico di Torino Published DOI: Terms of use: Altro tipo di accesso This article is made available under terms and conditions as specified in the corresponding bibliographic description in the repository Publisher copyright (Article begins on next page) 08 October 2021 Doctoral Dissertation Doctoral Program in Computer and Control Enginering (32nd cycle) Rethinking Software Network Data Planes in the Era of Microservices Sebastiano Miano ****** Supervisor Prof. Fulvio Risso Doctoral examination committee Prof. Antonio Barbalace, Referee, University of Edinburgh (UK) Prof. Costin Raiciu, Referee, Universitatea Politehnica Bucuresti (RO) Prof. Giuseppe Bianchi, University of Rome “Tor Vergata” (IT) Prof. Marco Chiesa, KTH Royal Institute of Technology (SE) Prof. Riccardo Sisto, Polytechnic University of Turin (IT) Politecnico di Torino 2020 This thesis is licensed under a Creative Commons License, Attribution - Noncommercial- NoDerivative Works 4.0 International: see www.creativecommons.org. The text may be reproduced for non-commercial purposes, provided that credit is given to the original author. I hereby declare that, the contents and organisation of this dissertation constitute my own original work and does not compromise in any way the rights of third parties, including those relating to the security of personal data. ........................................ Sebastiano Miano Turin, 2020 Summary With the advent of Software Defined Networks (SDN) and Network Functions Virtualization (NFV), software started playing a crucial role in the computer net- work architectures, with the end-hosts representing natural enforcement points for core network functionalities that go beyond simple switching and routing.
    [Show full text]
  • Data Warehousing on AWS
    Data Warehousing on AWS March 2016 Amazon Web Services – Data Warehousing on AWS March 2016 © 2016, Amazon Web Services, Inc. or its affiliates. All rights reserved. Notices This document is provided for informational purposes only. It represents AWS’s current product offerings and practices as of the date of issue of this document, which are subject to change without notice. Customers are responsible for making their own independent assessment of the information in this document and any use of AWS’s products or services, each of which is provided “as is” without warranty of any kind, whether express or implied. This document does not create any warranties, representations, contractual commitments, conditions or assurances from AWS, its affiliates, suppliers or licensors. The responsibilities and liabilities of AWS to its customers are controlled by AWS agreements, and this document is not part of, nor does it modify, any agreement between AWS and its customers. Page 2 of 26 Amazon Web Services – Data Warehousing on AWS March 2016 Contents Abstract 4 Introduction 4 Modern Analytics and Data Warehousing Architecture 6 Analytics Architecture 6 Data Warehouse Technology Options 12 Row-Oriented Databases 12 Column-Oriented Databases 13 Massively Parallel Processing Architectures 15 Amazon Redshift Deep Dive 15 Performance 15 Durability and Availability 16 Scalability and Elasticity 16 Interfaces 17 Security 17 Cost Model 18 Ideal Usage Patterns 18 Anti-Patterns 19 Migrating to Amazon Redshift 20 One-Step Migration 20 Two-Step Migration 20 Tools for Database Migration 21 Designing Data Warehousing Workflows 21 Conclusion 24 Further Reading 25 Page 3 of 26 Amazon Web Services – Data Warehousing on AWS March 2016 Abstract Data engineers, data analysts, and developers in enterprises across the globe are looking to migrate data warehousing to the cloud to increase performance and lower costs.
    [Show full text]
  • Master Data Simplified
    MASTER THE BEST IN MASTER DATA GOVERNANCE DATA SIMPLIFICATION & MANAGEMENT SIMPLIFIED CHARLIE MASSOGLIA & ANBARASAN MURUGAN ‘A MUST READ FOR DATA ENTHUISIASTS’ - AUSTIN DAVIS About the Authors Charlie Massoglia VP & CIO, Chain-Sys Corporation Former CIO for Dawn Food Products For 13+ years. 25+ years experience with a variety of ERP systems. Extensive experience in system migrations & conversions. Participated in 9 acquisitions ranging from a single US location to 14 sites in 11 countries. Author of numerous technical books, articles, presentations, and seminars globally. Anbarasan Murugan Product Lead, Master Data Management Master Data Simplification & Governance expert. Industry experience of more than 11 years. Chief Technical Architect for more than 10 products TM within the Chain Sys Platform . Has designed complex analytical & transactional Master data processes for Fortune 500 companies. Master Data Simplified An Introduction to Master Data Simplification, Governance, and Management By Charles L. Massoglia VP & CIO Chain●Sys Corporation [email protected] and Anbarasan Murugan Product Manager Chain●Sys Corporation [email protected] No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without written permission of the publisher. For information regarding permission, write to Chain-Sys Corporation, Attention: Permissions Department, 325 S. Clinton Street, Suite 205, Grand Ledge, MI 48837 Trademarks: Chain●Sys Platform is a trademark of Chain-Sys Corporation in the United States and other countries and may not be used without permission.
    [Show full text]
  • Lineage Tracing for General Data Warehouse Transformations
    Lineage Tracing for General Data Warehouse Transformations∗ Yingwei Cui and Jennifer Widom Computer Science Department, Stanford University fcyw, [email protected] Abstract. Data warehousing systems integrate information and managing such transformations as part of the extract- from operational data sources into a central repository to enable transform-load (ETL) process, e.g., [Inf, Mic, PPD, Sag]. analysis and mining of the integrated information. During the The transformations may vary from simple algebraic op- integration process, source data typically undergoes a series of erations or aggregations to complex procedural code. transformations, which may vary from simple algebraic opera- In this paper we consider the problem of lineage trac- tions or aggregations to complex “data cleansing” procedures. ing for data warehouses created by general transforma- In a warehousing environment, the data lineage problem is that tions. Since we no longer have the luxury of a fixed set of of tracing warehouse data items back to the original source items operators or the algebraic properties offered by relational from which they were derived. We formally define the lineage views, the problem is considerably more difficult and tracing problem in the presence of general data warehouse trans- open-ended than previous work on lineage tracing. Fur- formations, and we present algorithms for lineage tracing in this thermore, since transformation graphs in real ETL pro- environment. Our tracing procedures take advantage of known cesses can often be quite complex—containing as many structure or properties of transformations when present, but also as 60 or more transformations—the storage requirements work in the absence of such information.
    [Show full text]
  • Doppiodb 2.0: Hardware Techniques for Improved Integration of Machine Learning Into Databases
    Research Collection Conference Paper doppioDB 2.0: Hardware Techniques for Improved Integration of Machine Learning into Databases Author(s): Kara, Kaan; Wang, Zeke; Zhang, Ce; Alonso, Gustavo Publication Date: 2019-08 Permanent Link: https://doi.org/10.3929/ethz-b-000394510 Originally published in: Proceedings of the VLDB Endowment 12(12), http://doi.org/10.14778/3352063.3352074 Rights / License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use. ETH Library doppioDB 2.0: Hardware Techniques for Improved Integration of Machine Learning into Databases Kaan Kara Zeke Wang Ce Zhang Gustavo Alonso Systems Group, Department of Computer Science ETH Zurich, Switzerland fi[email protected] ABSTRACT t1 compressed/ Database engines are starting to incorporate machine learning (ML) doppioDB 2.0 encrypted functionality as part of their repertoire. Machine learning algo- Table t1 Iterative Decryption rithms, however, have very different characteristics than those of Execution Decompression relational operators. In this demonstration, we explore the chal- SCD lenges that arise when integrating generalized linear models into a t1 bitweaving t1_model database engine and how to incorporate hardware accelerators into Iterative Quantized Execution SGD the execution, a tool now widely used for ML workloads. t1_model The demo explores two complementary alternatives: (1) how to - Training: INSERT INTO t1_model train models directly on compressed/encrypted column-stores us- SELECT weights FROM TRAIN('t1', step_size, …); ing a specialized coordinate descent engine, and (2) how to use a - Validation: SELECT loss FROM VALIDATE('t1_model', 't1'); bitwise weaving index for stochastic gradient descent on low pre- SELECT prediction FROM INFER('t1_model', 't1_new'); cision input data.
    [Show full text]
  • Extract, Transform, Load | ETL Development
    What is ETL – Extract, Transform, Load | ETL Development ETL development – Pre requisites. 1. Setup Source and Target database. 2. Creation of ODBC connections. 1. Creating source ODBC connection. 2. Creating target ODBC connection. 3. Starting Informatica PowerCenter service. 4. Creating folder. ETL development process. 1. Creation of source metadata. 2. Creation of target metadata. 3. Design mapping without business rules. 4. Creating session for each mapping. 1. Create reader connection (source). 2. Create writer connection (Target). 3. Create workflow. 4. Run workflow. 5. Monitoring ETL process. (view state). What is ETL? ETL stands for Extract-Transform-Load. ETL is the process of extracting the data from different source (Operational databases) systems, integrating the data and Transforming the data into a homogeneous format and loading into the target warehouse database. Simply the overall process of ETL (Extraction, Transformation and Loading) is called Data Acquisition. Extraction : Extraction is the process of reading the data from source databases into staging areas. Transformation : Transformation is the process of converting the source data into required warehouse format. Loading : Loading is the process of writing converted data from staging area into target warehouse systems. What are the GUI based ETL tools? 1. Informatica. 2. DataStage. 3. Data Junction. 4. Oracle Warehouse Builder. 5. Cognos Decision Stream. What are the programmatic based ETL tools? 1. PI/Sql. 2. SAS Base. 3. Tera ACCESS. 4. Tera Data Utilities. 1. BTEQ. 2. Fast Load. 3. Multi Load. 4. T(Trickle) Pump. An Informatica PowerCenter is a GUI based ETL (Extract, Transform, Load) tool from Informatica Corporation which was founded in Redwood city, Los Angels (1993).
    [Show full text]
  • Rethinking Software Network Data Planes in the Era of Microservices
    Doctoral Dissertation Doctoral Program in Computer and Control Enginering (32nd cycle) Rethinking Software Network Data Planes in the Era of Microservices Sebastiano Miano ****** Supervisor Prof. Fulvio Risso Doctoral examination committee Prof. Antonio Barbalace, Referee, University of Edinburgh (UK) Prof. Costin Raiciu, Referee, Universitatea Politehnica Bucuresti (RO) Prof. Giuseppe Bianchi, University of Rome “Tor Vergata” (IT) Prof. Marco Chiesa, KTH Royal Institute of Technology (SE) Prof. Riccardo Sisto, Polytechnic University of Turin (IT) Politecnico di Torino 2020 This thesis is licensed under a Creative Commons License, Attribution - Noncommercial- NoDerivative Works 4.0 International: see www.creativecommons.org. The text may be reproduced for non-commercial purposes, provided that credit is given to the original author. I hereby declare that, the contents and organisation of this dissertation constitute my own original work and does not compromise in any way the rights of third parties, including those relating to the security of personal data. ........................................ Sebastiano Miano Turin, 2020 Summary With the advent of Software Defined Networks (SDN) and Network Functions Virtualization (NFV), software started playing a crucial role in the computer net- work architectures, with the end-hosts representing natural enforcement points for core network functionalities that go beyond simple switching and routing. Recently, there has been a definite shift in the paradigms used to develop and deploy server applications in favor of microservices, which has also brought a visible change in the type and requirements of network functionalities deployed across the data center. Network applications should be able to continuously adapt to the runtime behav- ior of cloud-native applications, which might regularly change or be scheduled by an orchestrator, or easily interact with existing “native” applications by leveraging kernel functionalities - all of this without sacrificing performance or flexibility.
    [Show full text]
  • Doppiodb 2.0: Hardware Techniques for Improved Integration of Machine Learning Into Databases
    doppioDB 2.0: Hardware Techniques for Improved Integration of Machine Learning into Databases Kaan Kara Zeke Wang Ce Zhang Gustavo Alonso Systems Group, Department of Computer Science ETH Zurich, Switzerland fi[email protected] ABSTRACT t1 compressed/ Database engines are starting to incorporate machine learning (ML) doppioDB 2.0 encrypted functionality as part of their repertoire. Machine learning algo- Table t1 Iterative Decryption rithms, however, have very different characteristics than those of Execution Decompression relational operators. In this demonstration, we explore the chal- SCD lenges that arise when integrating generalized linear models into a t1 bitweaving t1_model database engine and how to incorporate hardware accelerators into Iterative Quantized Execution SGD the execution, a tool now widely used for ML workloads. t1_model The demo explores two complementary alternatives: (1) how to - Training: INSERT INTO t1_model train models directly on compressed/encrypted column-stores us- SELECT weights FROM TRAIN('t1', step_size, …); ing a specialized coordinate descent engine, and (2) how to use a - Validation: SELECT loss FROM VALIDATE('t1_model', 't1'); bitwise weaving index for stochastic gradient descent on low pre- SELECT prediction FROM INFER('t1_model', 't1_new'); cision input data. We present these techniques as implemented in - Inference: our prototype database doppioDB 2.0 and show how the new func- tionality can be used from SQL. Figure 1: Overview of an ML workflow in doppioDB 2.0. PVLDB Reference Format: Kaan Kara, Zeke Wang, Ce Zhang, Gustavo Alonso. doppioDB 2.0: Hard- and compress data for better memory bandwidth utilization and de- ware Techniques for Improved Integration of Machine Learning into Databases.
    [Show full text]
  • Automating the Capture of Data Transformation Metadata
    Automating the Capture of Data Transformation Metadata H.V. Jagadish Univ. of Michigan http://www.eecs.umich.edu/~jag George Alter, University of Michigan Why Metadata? • Data are useless without Metadata – “data about data” • Metadata should: – Include all information about data creation – Describe transformations to variables – Be easy to create • Our goal: Automated capture of metadata A few words about ICPSR • World’s largest archive of social science data • Consortium established 1962 • 760+ member institutions around the world • Founding member and home office for the DDI Alliance Powered by DDI Metadata ICPSR is building search tools based upon Data Documentation Initiative (DDI) XML Codebooks (pdf and online) are rendered from the DDI. Searchable database of 4.5M variables Click here for online codebook What question Online codebook shows was asked? variable in context of dataset How was the Link to online question coded? graph tool Link to online crosstab tool Searchable database of 4.5M variables Click here for variable comparison Variable comparison display Click here for online codebook Metadata for the American National Election Study What question Who answered was asked? this question? How was the question coded? Who answered this question? Metadata for the American National Election Study Who answered this question? How do we know who answered the question? It’s in the pdf. Who answered this question? When data arrive at the archive… • No question text • No interview flow (question order, skip pattern) • No variable provenance • Data transformations are not documented. How is research data created? • Most surveys are conducted with computer assisted interview software (CAI) – CATI – Computer-assisted Telephone Interview – CAPI – Computer-assisted Personal Interview – CAWI – Computer Aided Web Interview • There is no paper questionnaire • The CAI program is the questionnaire – i.e.
    [Show full text]