#Fuse spreadsheet corpus tools

This repository contains all the software used to extract the Fuse corpus from CommonCrawl. It is primarily designed to be run in a Hadoop environment, however, the spreadsheet-analyzer project has been configured to run on any Java 1.7+ environment.

###Build

Dependencies (and the rest of the build process) are managed by Maven. To build:

1. Install Maven.
2. Clone repository. Navigate a command prompt to the folder.
3. Run mvn clean install, which will build and install all three projects.

Useful Maven Commands:

• Run mvn clean install from parent directory to do a full build.
• Run mvn eclipse:eclipse to build an Eclipse project for one of the modules.
• Run mvn eclipse:eclipse -DdownloadSources=true -DdownloadJavadocs=true to build an Eclipse project with javadocs and sources linked.
• Run mvn clean compile assembly:single in one of the module folders to build a jar with dependencies included source

####About the projects

• spreadsheet-crawler : contains many tasks useful for extracting spreadsheets from CommonCrawl.

• spreadsheet-analyzer : A standalone analyzer which uses a slightly modified version of Apache POI (located under custom-poi) to generate summary analysis of a directory of spreadsheets. After being compiled with all its dependencies, it can be run via java -cp analyzer-with-dependencies.jar net.barik.spreadsheet.analysis.JSONScanner dir/to/scan/

• spreadsheet-anlyzer : A wrapper for spreadsheet-analyzer that allows it to be run in a Hadoop map-reduce environment.

####Example Configurations for running in AWS EMR

We have found some tweaks to the EMR configurations help the jobs complete faster and we list those here:

Extracting spreadsheets from Common Crawl:

• Hadoop configuration: -m mapred.map.tasks.speculative.execution=false -m mapred.reduce.tasks=0 -c fs.s3n.ssl.enabled=false -m mapreduce.map.java.opts=-Xmx4096m -m mapreduce.map.memory.mb=4096 -m io.file.buffer.size=65536 -m mapreduce.task.timeout=1200000 -y yarn.scheduler.maximum-allocation-mb=4096 -y yarn.nodemanager.resource.cpu-vcores=1 -y yarn.scheduler.minimum-allocation-mb=4096
• Custom Jar arguments: net.barik.spreadsheet.WATExtractJob -D export.bucket="out-bucket" -D export.keyprefix=output/dir/ s3n://location/to/wat/path s3n://location/for/logs

Analyzing spreadsheets:

• Hadoop configuration: -m mapred.map.tasks.speculative.execution=false -m mapred.reduce.tasks=0 -c fs.s3n.ssl.enabled=false -m io.file.buffer.size=65536 -m mapreduce.task.timeout=2400000 -m mapreduce.map.memory.mb=5500 -m mapreduce.map.java.opts=-Xmx5200m -y yarn.scheduler.maximum-allocation-mb=6000 -y yarn.nodemanager.resource.cpu-vcores=1 -y yarn.nodemanager.resource.memory-mb=7000
• Custom Jar arguments: -D import.bucket="myBucket" -D export.bucket="myBucket" -D import.keyprefix=dir/with/spreadsheets/ -D export.keyprefix=analysis/output/ -D corpus.name=Fuse s3n://location/to/list/of/spreadsheets s3n://location/for/logs

###License We modified the 3.11 version of Apache POI to allow it to deal with slightly malformed spreadsheets. We include the compiled, modified jars in this repo (see NOTICE). All software is released under the BSD license (see LICENSE).