Data transformation workflows - Wolfram U sessions

Ideas
, , , , , ,
1

Data transformation workflows

General description

The Data Transformation (DT) lectures outlined below are to be recorded for
Wolfram U
in October-December 2020.

  • Here is the Wolfram U recordings playlist at YouTube

  • Each tutorial lecture is between 25 and 40 minutes.

  • Most of the workflows in the lectures have (easy to derive) corresponding code in the programming languages: Julia, Python, R, and WL.
    (In Python and Julia dedicated data transformation packages are utilized.)

  • For didactic purposes we are going to use the Domain Specific Language (DSL)
    Raku-programmed
    system
    DSL::English::DataQueryWorkflows, [AAp1],
    that generates code for the programming languages R and WL and
    the software packages:

  • The WL generated code uses functions from the Wolfram Function Repository.
    (See the references section “Wolfram Function Repository functions”.)

  • The tutorials sequence has an “inverted” structure: the most conceptually rich and technical sessions are presented first.

Detailed tutorial plan

Introduction through questions and answers

  • What are the simplifying assumptions for the family of workflows considered in this tutorial?

    • We do not consider hard to generalize- or too specialized transformations.
    • For example, we do not consider:
      • Data gathering/harvesting
      • Data ingestion by parsing of, say, XML structures
      • Date-time specific transformations
      • Natural text specific transformations
      • Etc.

  • What are the simplifying assumptions about the targeted data and transformations?

    • Tabular data and collections of tabular data. (E.g. lists of datasets.)
    • Transformation workflows that can be expressed with a certain “standard” or “well-known” subset of SQL.
    • The flow chart given below illustrates well the targeted DT.

  • Do these workflows apply to other programming languages and/or data systems?

    • Yes, with the presented DT know-how we target multiple “data science” programming languages.

  • Are additional packages needed to run the codes?

    • Yes and no: depends on the target data transformation system.
    • Sometimes the WL code uses Wolfram Function Repository functions.

  • Are there similar tutorials dedicated to other programming languages or packages?

    • Yes, many. Both for WL and the rest (Julia/Python/R.)
    • That said, in this tutorial we use a certain simplified and streamlined data-and-transformations model
      that allows the development of cross-system, transferable know-how and workflows.

  • What are the most important concepts for a newcomer to data wrangling?

    1. Cross tabulation (or contingency matrices)
    2. (Inner) joins
    3. The so called Split-apply-combine pattern
    4. Long form and wide form (or data pivoting)

  • How do the considered DT workflows relate to well known Machine Learning (ML) workflows?

    • Generally speaking, data has to be transformed into formats convenient for the application of ML algorithms.
      How much transformation is needed depends a lot on the host system or targeted ML package(s).
    • For example, WL has extensive ML data on-boarding functions.
      (“Encoders” and “decoders” that make the use of Neural Networks more streamlined.)
      Hence, for using ML in WL less DT is needed.

  • How much is WL used?

    • Approximately 80% we use WL; for illustration purposes we show some of the workflows in other languages.
      (Run within the same Mathematica presentation notebooks.)

  • Mental model (of the considered DT queries)

    • Simplifying assumptions – targeting:
      • Tabular data
      • Collections of tabular data
    • Workflows considered:
      • (Generalizing) flow chart
      • The Split-transform-combine DT pattern, [HW1]
      • Reshaping

Reshaping of data – detailed examples

  • Long form (aka “narrow format”):

    • Using long form vs not using long form
    • Programmatic use
    • Relation to sparse matrices

  • Wide form (aka “wide format”)

    • As an inversion of long form
    • Typical usage

  • Cross tabulation

    • Relation to wide form
    • The different missions and mind-share of cross tabulation and wide form

Basic operations and joins

  • Basic operations

    • Data load
    • Column selection
    • Row filtering
    • Summarization
      • Ways to summarize data
      • “Global” summary
      • Summarization per group and per column
      • Summarization vs Cross tabulation

  • Joins

    • Types of joins
    • Inner joins
    • Left joins
    • Semi-joins

Additional topics

  • Example combinations with ML workflows
    • Making a recommender system
    • Making a classifier

Visual aids

Here is a flow chart that encompasses a large fraction of typical tabular data transformation workflows:

TabularDataTransformationWorkflow
TabularDataTransformationWorkflow2318Ă—1024 145 KB

This call of
ToDataQueryWorkflow
through
ExternalParsersHookUp, [AAp2],
over a sequence of natural commands:

ToDataQueryWorkflowCode["use dfTitanic; 
      filter with passengerSurvival is 'survived';
      rename passengerClass, passengerSex as class and sex;
      cross tabulate class and sex", "Execute" -> False]

generates this WL code:

Hold[obj = dfTitanic; 
 obj = Select[obj, #1["passengerSurvival"] == "survived" &]; 
 obj = (Join[KeyDrop[#1, {"passengerClass", "passengerSex"}], Association["class" -> #1["passengerClass"], "sex" -> #1["passengerSex"]]] &) /@ obj; 
 obj = ResourceFunction["CrossTabulate"][({#1["class"], #1["sex"]} &) /@ obj]]

which when executed gives the dataset:

Dataset[<|
"1st" -> <|"female" -> 139, "male" -> 61|>, 
"2nd" -> <|"female" -> 94, "male" -> 25|>, 
"3rd" -> <|"female" -> 106, "male" -> 75|>|>, 
 TypeSystem`Assoc[TypeSystem`Atom[String], TypeSystem`Struct[{"female", "male"}, {TypeSystem`Atom[Integer], TypeSystem`Atom[Integer]}], 3], 
 <|"ID" -> 177940708079520|>]

References

Articles

[HW1] Hadley Wickham,
“The Split-Apply-Combine Strategy for Data Analysis”,
(2011),
Journal of Statistical Software.

[AA1] Anton Antonov,
“Contingency tables creation examples”,
(2016),
MathematicaForPrediction at WordPress.

[AA2] Anton Antonov,
“Pareto principle adherence examples”,
(2016),
MathematicaForPrediction at WordPress.

Packages

[AAp1] Anton Antonov,
DSL::English::DataQueryWorkflows Raku package,
(2020),
GitHub/antononcube.

[AAp2] Anton Antonov,
External Parsers Hookup Mathematica package,
(2020),
ConversationalAgents at GitHub.

[AAp3] Anton Antonov,
Cross tabulation implementation in Mathematica,
(2017),
MathematicaForPrediction at GitHub.

[AAp4] Anton Antonov,
Data reshaping Mathematica package,
(2018),
MathematicaForPrediction at GitHub.

Wolfram Function Repository functions

Anton Antonov, CrossTabulate.

Anton Antonov, ExampleDataset.
(Submitted)

Anton Antonov, ParetoPrinciplePlot.

Anton Antonov, RecordsSummary.

Anton Antonov, ToLongForm.
(Submitted)

Anton Antonov, ToWideForm.
(Submitted)

Seth Chandler, ResetDataset.

Data Transformations Workflows Q_AMP_A Introduction.nb (1.23 MB)

4 Likes