FSharp and Data 101

F# and Data 101Utilizing FsLab to Quickly Understand your Data

Jonathan Dexter, Technology Manager of .NET, The Nerdery

AgendaWhat is this talk aboutGet the data!Transform!Science!???Prot!

If you came here to hear about"Monads""Functors"Tail-call optimizationImmutable designPattern matching

What we WILL talk about : FsLab

With Paket

1: 2:

paketinitpaketaddnugetfslab

With NuGet

1: nugetinstallfslabOutputDirectorypackages

The process(Not limited to FsLab)

1: 2: 3: 4:

acquiredata|>transform|>science|>visualize

FsLab: Scratching the Surface5 Libraries

Sorting the libraries above into categories:

Library Acquire Transform Science Display

F# Data Yep! - - -

Deedle - Yep - -

.NET Numerics - Supports Supports -

R Type Provider Partial Yep Yep Partial

XPlot - - - Yep

Step one: Acquire

Classic scenario: CSV

Using CSV type provider

1: 2:

typecsv=FSharp.Data.CsvProviderletcomplaints=csv.Load(complaintsCsv)

Using a data frame

1: letdata=Deedle.Frame.ReadCsv(__SOURCE_DIRECTORY__+"/data.csv")

Type Provider Scenario: World bank provider

World bank provider is bundled with F# Data

1: 2: 3: 4: 5: 6:

letdataContext=FSharp.Data.WorldBankData.GetDataContext()lethighTechExports=dataContext.Countries.``UnitedStates``.Indicators.``Hightechnologyexports(currentUS$)``

Type Provider Scenario: JSON provider

1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20:

typeJsonContext=FSharp.Data.JsonProvider

Type Provider Scenario: JSON Provider (cont.)

1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12:

letmissedVotes=JsonContext.Load(sprintf"%s?apikey=%s"missedVotesUrlapikey)

letcongressmen=missedVotes.Results|>Seq.collect(funr>r.Members)

lettopMissingCongressman=congressmen|>Seq.sortBy(funm>trym.MissedVotesPctwith|ex>0.0m)|>Seq.rev|>Seq.head

Type Provider Scenario: JSON Provider (cont.)

{"id":"M000309","name":"CarolynMcCarthy","party":"D","state":"NY","district":"4","total_votes":"1192","missed_votes":"687","missed_votes_pct":"57.63","rank":"1","notes":"Willretireattheendof113thCongress."}

Batteries not included: SQL Provider

*SQL Data Connection

*SQL Entity Connectoin

SQL Client

SQL Provider

https://msdn.microsoft.com/en-us/library/hh362320.aspxhttps://msdn.microsoft.com/en-us/library/hh362320.aspxhttps://github.com/fsprojects/FSharp.Data.SqlClienthttps://github.com/fsprojects/SQLProvider

SQL Data Connection

Type provder for an entire database, MS SQL focused.

1: 2: 3: 4:

typedbSchema=SqlDataConnectionletdb=dbSchema.GetDataContext()

SQL Entity Connection

Type provder for an entire database, through ADO.NET Entity model.

1: 2: 3: 4:

typedbSchema=SqlEntityConnectionletdb=dbSchema.GetDataContext()

SQL Client

Type provider for commands, sprocs, and queries

1: 2: 3: 4:

usecmd=newSqlCommandProvider()

1: letresults=cmd.Execute(region="USA")

SQL Provider

Type provider for DB as a whole

MS SQL, Postgres, SQLite, MySQL, Oracle, MS Access

1: 2: 3: 4:

typesql=SqlDataProvider

letctx=sql.GetDataContext()

Step two: Transform

Deedle: Convert to data frame

1: 2: 3:

openDeedleopenFSharp.DataopenFSharp.Data.Runtime.BaseTypes

Expander code omitted, but can be found here

1: 2: 3: 4:

letdataFrame=[forlincongressmen>series["It"=>l]]|>Frame.ofRowsOrdinal|>Frame.expandAllCols10

https://github.com/fslaborg/FsLab/issues/14

Deedle: Normal syntax

1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12:

lethighTechExportData=WorldBankData.GetDataContext().Countries.``UnitedStates``.Indicators.``Hightechnologyexports(currentUS$)``

lethighTechFrame=highTechExportData|>Frame.ofRecords|>Frame.indexRowsInt"Item1"|>Frame.mapColKeys(fun_>"HighTechExports")

Deedle: Quick manipulations

Simple statistics

1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15:

letstats=["Min"=>Stats.minhighTechFrame"Max"=>Stats.maxhighTechFrame"Average"=>Stats.meanhighTechFrame"StandardDeviation"=>Stats.stdDevhighTechFrame]letobservations=highTechFrame?``HighTechExports``|>Series.observations|>Seq.map(fun(k,v)>floatk,floatv)

letregression=observations|>MathNet.Numerics.LinearRegression.SimpleRegression.Fit

Results

Stats

[("Min",series[HighTechExports=>76767867475])("Max",series[HighTechExports=>220884471208])("Average",series[HighTechExports=>152642394565.462])("StandardDeviation",series[HighTechExports=>39746534154.5001])]

Regression fit (intercept, slope)

(5.799301011e+12,2973741397.0)

R Squared

R squared value

1: 2:

letrsquared=GoodnessOfFit.RSquared(regressedValues|>Seq.mapsnd,observations|>Seq.mapsnd)

0.3274641292

Deedle: Combining data and additional feature creation

1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14:

letexportFrame=WorldBankData.GetDataContext().Countries.``UnitedStates``.Indicators.``Exportsofgoodsandservices(currentUS$)``|>Frame.ofRecords|>Frame.indexRowsInt"Item1"|>Frame.mapColKeys(fun_>"TotalExports")

exportFrame?``HighTechExports``

Deedle: Straight to R

Arrrr

Deedle: Straight to R

1: 2: 3: 4:

openRProvider.``base``openRProviderletrFrame=R.as_data_frame(exportFrame)letrFrameSummary=(R.summaryrFrame)

["TotalExportsMin.:2.700e+10""TotalExports1stQu.:1.110e+11""TotalExportsMedian:3.639e+11""TotalExportsMean:6.684e+11""TotalExports3rdQu.:1.015e+12""TotalExportsMax.:2.342e+12""HighTechExportsMin.:7.677e+10""HighTechExports1stQu.:1.282e+11""HighTechExportsMedian:1.521e+11""HighTechExportsMean:1.526e+11""HighTechExports3rdQu.:1.763e+11""HighTechExportsMax.:2.209e+11""HighTechExportsNA's:29""PercentageofHighTechExportsMin.:0.06563""PercentageofHighTechExports1stQu.:0.12270""PercentageofHighTechExportsMedian:0.15502""PercentageofHighTechExportsMean:0.13875""PercentageofHighTechExports3rdQu.:0.16385""PercentageofHighTechExportsMax.:0.18290""PercentageofHighTechExportsNA's:29"]

Step three: Science

"Normal" data analysis languages

F# is slowly catching up

Machine Learning Algorithms (suite)


Natural Language Processing


Cloud computing

Step four: Visualize

Visualizing our previous information

With Google Charts

1: 2: 3: 4: 5: 6:

letpieChart=congressmen|>Seq.filter(func>tryc.MissedVotes>=0with_>false)|>Seq.groupBy(func>c.Party)|>Seq.map(fung>fstg,(sndg)|>Seq.sumBy(func>c.MissedVotes|>XPlot.GoogleCharts.Chart.Pie


1: 2: 3: 4:

letscatter=exportFrame?``TotalExports``|>Series.observations|>XPlot.GoogleCharts.Chart.Scatter

Summary

ResourcesPresentation code:

Presentation share:https://github.com/mandest/FSharpAndDataTalk

More F# Resources: *F# Guides on fsharp.org Functional Programming SlackF# Weekly

https://github.com/mandest/FSharpAndDataTalkhttp://fsharp.org/http://fpchat.com/https://sergeytihon.wordpress.com/category/f-weekly/