Choose Language Hide Translation Bar

Functional DOE: Tips and Tricks for Design and Organizing Data (2022-US-45MP-1152)

The Functional Data Explorer (FDE) in JMP Pro allows for analysis of a DOE where the response is a curve. The entire functional DOE analysis workflow can be done within FDE - from smoothing response curves all the way to fitting the functional DOE model and optimization with the Profiler. But what about setting up the design and organizing your data for functional DOE analysis? This presentation will help you understand options around functional DOE design using the Custom Design platform and organizing your data using table manipulations such as Stack, Split, Join, and Update.

 

 

Hi,  everyone. My  name  is  Andrea  Coombs.

I'm  a  Senior  Systems  Engineer, supporting  customers

from  major  accounts in  the  eastern  part  of  the  US.

Today  I'm  going  to  be  talking  about functional DOE

and  specifically  around  design of  your  functional  DOEs

and  how  to  prepare   your  data  for  analysis.

I'm  going  to  turn  off my  video  for  presentation  here.

Let's  go  in  and  look  at  the  goals.

Really,  the  goals  are  very  simple here.

I  want  to  cover  some  tips  and  tricks for  setting  up  your  functional  DOE

using  the  Custom  Design  Platform

and  also  give  you  some  tips  and  tricks

for  adding  functional  data to  your  DOE  data  table.

I  am  going  to  be  using  J MP Pro  16.2 during  this  presentation.

Let's  start  off  by  defining what  is  functional  data,

what  is  functional  data  analysis, and  specifically,  what  is  functional  DOE.

Functional  data  is  really  curved  data.

It's  any  data  that   unfolds  over  a  continuum,

and  there's  a  lot  of  data that  is  inherently  functional  in form.

You  can  think  about  time  series  data,

sensor  streams from  a  manufacturing  process,

spectra  data  that's  produced by  lots  of  different  types  of  equipment,

measurements  taken   over  a  range  of  temperatures.

And  the  Functional  Data  Explorer in  JMP  Pro  makes  it  really  easy

to  solve  many  kinds  of  problems with  functional  data.

Here  we  have  an  example of  some  functional  data.

Here  we  have  a  plot  of  Home  Price  Index for  New  Jersey,  from  1990  to  2021.

Llike  many  functional  data,

we  don't  get  it  as  this  smooth  curve, like  we  see  here,

but  rather  we  get a  series  of  discrete  Index  values.

So  we  get  one  value  that  represents the  value  for  the  X,  the  year,

and  the  value  for  Y,  the  Home  Price  Index.

With  functional  data  analysis,

it  isn't  typically  just  one  point of  the  curve  that  we  are  interested,

or  even  the  collection of  points  from  a  single  curve.

We  typically  have  a  collection  of  curves.

Here  you  can  see  we  have the  Home  Price  Index  over  time,

for  the  50  states, plus  the  District  of  Columbia.

And  when  we're  doing functional  data  analysis,

we  want  to  understand the  variability  around  these  curves.

Often  we  want  to  understand what  are  the  variables

that  drive  the  variability  in  our  curves.

Or  maybe  you  want  to  use  the  variability in  our  curves  to  predict  another  outcome.

Functional  data  analysis

is  going  to  use  all of  the  information contained  in  the  curves.

We're  not  going  to  leave any  information  behind.

To  model  the  curves  directly,

we  can  treat  the  curves   as  first  class  data  objects,

in  the  same  way  that  JMP  will  treat traditional  types  of  scalar  data.

When  I'm  thinking  about functional  data  analysis,

I  like  to  break  this  down  into  four  steps.

The  first  step  is  to  take the  collection  of  curves

and  to  smooth  the  individual  curves.

The  next  thing  is  we'll  determine  the  mean of  those  curves  and  the  shape  components.

These  shape  components  represent the  variability  around  the  mean.

Next,  we  extract  the  magnitude of  each  of  these  shape  components.

Knowing  the  magnitude of  the  shape  components,

the  function  that  describes the  shape  component,

the  function  that  describes  the  mean, we  are  able  to  reproduce  all  of  our  curves

by  just  knowing  these  two  shape  component  scores.

Now  we  can  use  the  shape   component  scores  in  an  analysis.

Here  what  I've  done is  I've  done  a  cluster  analysis,

and  I've  defined four  groups  of  my  curve  shapes.

In  the  Functional  Data  Explorer  itself,

there  are  two  primary  questions that  can  be  answered.

The  first  is  about  how  to  adjust process  settings  and  product  inputs

to  achieve  a  desired   function  or  spectral  shape.

We  call  this  Functional  DOE  analysis, or  FunDoE,  for  short.

The  second  question  we  can  ask  is, how  can  I  use  entire  functions

to  predict  something  like  yield or  quality  attributes?

We  call  this  Functional m achine  learning or  FunML,  for  short.

Today  we're  going  to  be focusing  on  Functional  DOE.

Let's  take  a  closer  look at  the  Functional  DOE  workflow.

The  first  thing  you  want  to  do is  to  set up  your  design

using  the  Custom  Design  Platform.

Then  you  can  go  out,  run  your  DOE,

collect  the  results, and  you  want  to  organize  your  data

to  get  it  ready  to  put  into  the Functional  Data  Explorer.

The  remaining  steps in  our  Functional  DOE  workflow,

will  be  done  all  within the  Functional  DOE  Explorer.

In  the  Functional  Data  Explorer, we  can  process  our  data,

we  can  smooth  our  individual  curves, we  can  extract  our  shape  components,

and  then  we'll  use  our shape  component  scores  for  DOE  modeling,

and  we  can  use  our  profiler to  address  the  goals  of  our  DOE.

All  of  this  is  done  within the  Functional  Data  Explorer.

Now,  there  are  many  presentations at  this  Discovery  Summit,

at  previous  Discovery  Summit,

even  in  our  Mastering  JMP  series on  jmp.com,

that  will  go  over  lots  of  details  around the  Functional  Data  Explorer.

I'm  not  going  to  be  talking  specifically about  the  Functional  Data  Explorer  today.

What  I  want  to  talk  about  is how  do  you  set up  your  design

for  a  functional  DOE and  how  can  you  organize  your  data.

To  do  this,  I'm  going  to  use this  Bead  Mill  Case  Study.

In  this  example,  what  we  have  is

we're  essentially  milling pigment  particles  for  LED  screens.

You  start  off  with  beads  and  pigment in  this  slurry,  in  this  holding  tank.

It  goes  through, it  flows  through  this  milling  chamber,

and  comes  back  to  the holding  tank  in  a  continuous  process.

So  if  we  were  doing  a  DOE  on  this  process, some  factors  that  we  could  look  at

is  the  percent  of  beads  we're  starting  off with  here  in  the  holding  tank,

the  percent  of   pigment  particles we're  starting  off  with.

We  can  look  at  the  flow  rate through  the  system

and  also  the  temperature.

When  we're  looking at  the  goal  of  this  DOE,

we  essentially  want  to  achieve an  optimal  size  over  time  curve.

So  let's  take  a  look at  that  optimal  curve.

The  optimal  curve  is  represented by  this  green  curve  here.

So  essentially,  we  want our  pigment  particle  sizes  to  decrease,

so  they  fall  within  specification  quickly.

And  our  specification  range  is represented  by  this  green  shaded  area.

We  want  those  particles  to  remain within  specification

throughout  the  duration  of  the  run.

That  is  our  optimal  curve.

Let's  go  ahead and  take  a  look  at  data  prep.

I'm  going  to  talk  about  data  prep  first,

and  then  we'll  move  backwards and  talk  about  the  DOE  design.

For  data  prep,  there  are  three  main tips  and  tricks  I  want  to  share  with  you.

First  of  all,  I  want  you  to  understand

that  the  Functional  Data  Explorer accepts  data  in  different  formats.

The  Stacked  Data  Format  is  the  default format,  and  it's  the  most  versatile.

But  you  can  also  use  Rows  as  Functions.

I'm  going  to  go  over some  table  manipulations,

such  as  Stack,  Split,  Join,  and  Update,

to  show  you  how  you  can get  your  data  ready  for  analysis.

And  then  I'll  also  show  you how  you  can  quickly  import  multiple  files

if  your  curved  data  is  stored in  separate  files.

What  data  format  is  FDE  expecting?

Well,  there's  actually three  different  formats.

There's  Rows  as Functions, Stacked  Data  and  Columns as Functions.

Let  me  open  up  a  data  table  here and  launch  the  FDE  platform

to  show  you  that  there  are  different  tabs up  here  for  these  different  formats.

The  Stacked  Data  format  is  the  default.

We  have  Rows as  Functions and  Columns  as  Functions.

This  example  here  happens to  be  Rows  as Function.

Each  row  contains  a  full  function.

Here  we  have  the  first  run  from  our DOE,

and  the  function  is represented  here  in  these  columns.

Each  column  represents an  X  variable  or  an  input,

and  then  the  value  within  the  cell is  a  Y  variable.

When  we  go  to  populate the  Functional  Data  Explorer,

we  can  come  in  here, go  to  Rows  as Functions,

our  Y  output  is  represented in  these  columns,

we  can  put  in  our  DOE  factors,

and our  ID  function,  and  then  you  can go  ahead  and  analyze  that  data.

So  this  is  Rows as  Functions.

One  thing  to  know  is  that  Rows  as  Function

assumes  that  observations are  equally  spaced  in  the  input  domain,

unless  you  have  an  FDE X  Column  Property.

The  FDE X  Column  Property   is  something  that  comes  into  play

when  we  design  our DOE,

which  we're  going  to  talk about  here  in  a  second.

But  I  just  want  to  show  you  here,

next  to  each  of  these  columns,

I  have  a  Column  Property   associated  with  it.

And  you  can  see

here's  the  FDE X  Column  Property, and  the  X  input  value  will  be  two  here.

If  you  want  to  use the   FDE X Column Property ,

I'll  show  you  here  at  the  end

how  you  can  use  the  JMP  scripting  language  to  assign  that.

So  that's   Rows as Functions.

Now  let's  look  at  Stacked  Data.

Here's  an  example  of  Stacked  Data,

where  I  have  one  run  or  one  curve over  multiple  rows,

and  each  row  is an  observation  of  the  curve.

So  in  row  one  here,

I  have  a  value  for  X  and  a  value  for  Y, and  that  continues  over  multiple  rows.

This  is  the  most  common and  the  most  versatile  way

of  organizing  your   curve data.

And  when  we  populate the  Functional  Data  Explorer,

we're  here  in  the  Stacked  Data  format,

we'll  put  in  our  X and  our  Y  of  our  function,

put  in  our  ID  of  the  function,

and  then  we  can  put  in  our  DOE  Factors here  as  supplementary  variables.

The  last  type  of  format  that  the  functional  data  can  use

is  Columns  as  Functions.

I've  never  seen  data  organized  this  way, and  it's  a  little  perplexing.

It's  hard  for  me  to  get  my  mind  around why  you  would  organize  your  data  this  way,

but  I'll  show  it  to  you  even  though it's  not  very  common.

In  this  example,  each  row  is  the  level of  your  X  variable  of  the  function.

So  here  we  have  a  column  for  time,

and  each  row  represents  the  X measurement, and  then  each  column  represents  a  run.

Let's  go  ahead  and  launch the  Functional  Data  Explorer.

We'll  come  over  here to  Columns  as  Functions.

We  can  put  in  our  X  variable,

which  is  time,  and  all  of  our  output variables,  which  are  each  of  our  batches.

And  you'll  notice  in  here  we  cannot  input supplementary  variables

because  we  don't  have  any  way  of

defining  which  factor  or  which  treatment is  associated  with  each  of  these  runs.

So  you  cannot  do  Functional  DOE with  Columns  as  Functions.

Now  let's  talk  about  getting your  data  into  your  DOE  data  table.

To  do  this,  we're  going to  use  the  Tables  menu.

We  have  lots  of  different  platforms here where  we  can  manipulate  our  data.

The  two  things  that  we  may   want  to  do  with  our  data

is  reshape  it,  using  Sort  or  Stack,

or  we  may  want  to  add  data, by  using  Join,  Update,  or  Concatenate.

And  especially  if  you're  new  to  JMP, some  of  these  table  manipulation  platforms

can  be  a  little  confusing when  you  start  using  them.

The  little  icons next  to  each  of  the  platforms,

can  be  very  helpful to  know  which  platform  does  what.

So  what  I've  done  in  my  journal

is  I've  taken  each  of  these  icons and  I've  blown  them  up  here,

so  we  can  take  a  closer  look at  these  icons

to  understand  what each  of  these  platform  does.

Let's  first  talk  about reshaping  with  Stack  and  Split.

Let's  first  talk  about  Stacking.

Stacking  is  going  from  wide  to  tall  data.

In  this  example, you  have  data  in  multiple  columns,

and  you  want  to  combine that  data  into  one  column.

Let's  look  at  an  example  here.

Here  we  have  wide  data,

we  have  Rows as  Functions.

Let's  say  we  want  to  Stack  this  data,

so we  can  use  it in  the  Stacked  data  format,

in  the  Functional  Data  Explorer.

We're  going  to  come  up  here to  the  Tables  menu,  go  to  Stack.

I'm  going  to  pick  all  of  those columns  I  want  to  stack.

And  here  I  have  50  measurements in  each  of  my  functions.

I'm  going  to  select  all  50  of  those rows  and  say  I  want  to  stack  them.

I  can  come  down  here  and  define  what my  new  column  names  are  going  to  be.

The  data  that  I'm  stacking is  actually  my  size  data.

My  label  column,  which  happens to  be  my  column  name  here,

this  refers  to  my  time.

Now,  two  things  when  I'm  doing table  manipulation,

I  always  give  my  output table  an  explicit  name.

Otherwise,  JMP  will  call  it  Untitled,

and  it  will  iterate through  untitled  numbers.

So  I  like  to  give  them,

each  of  my  tables,   an  explicit  name,

and  then  keeping  dialog  open.

You  can  check  this  box to  keep  this  dialog  open,

so  when  you  hit  Apply and  see  your  results,

if  you  didn't  get  the  results you're  expecting,

you  have  your  dialog  here  to  review  what  you  did

and  maybe  fix  what  you  need  to  fix to  get  the  desired  output.

Now,  this  data  is  stacked  and  ready  to  go.

Let's  go  through  an  example  of  Split.

Split  is  when  you're  starting with  tall  data  or  stacked  data,

and  you  want t o  split  it  out into  different  columns.

In  this  example  here,  I  have  stacked  data,

and  let's  say  I  want  to  split  it  out so  I  can  use  Rows as  Functions  in  FDE.

I'm  going  to  come  up here  to  the  Tables  menu.

I'm  going  to  use  Split.

And  this  Split  dialog  is  probably the  most  confusing  of  all  of  them.

Even  after  using  JMP  for  many, many  years,

I  always  have  to   step  back and  think  about  how  to  populate  this.

But  the  Split  by  Columns

is  essentially  what's  going  to  be  your new  column  headers.

So  I  want  Time  as  my  new  column  headers, and  I  want  to  Split  out  my  size  data,

and  I  want  to  be  sure to  group  this  by  Run O rder.

A gain,  give  this  an  explicit  name,

and  I  can  keep  the  dialog open  to  see  how  I  split  this  data.

Now  here  I  have  my  data  is  wide, can  use  Rows as  Functions  in  the  FDE.

That  is  reshaping  your  data.

Going  from  wide  to  tall or  from  tall  to  wide.

Now  let's  talk  about  adding  data.

A  lot  of  times  you're  starting  out with  a  DOE  data  table  that  you  created,

such  as  this.

Let  me  just  delete  this  column  out.

I want  to  add,  I  want  to  be  able to  join  my   curve data  to  this  table.

Here's  my   curve data in  a  separate  table.

Essentially,  what  I  want  to  do is,

want  to  add  the  columns in  the  second  table  to  my  first  table.

I'm  going  to  use  Join. Join  adds  columns.

I'm  going  to  start here  with  my  DOE  table.

I'm  always  going  to  start with  my  DOE  table

because  my  DOE  table  has all  of  these  scripts  in  here

that  I  can  use  to  analyze  my  data.

These  are  very  important.

So you  always  want  to  start with  your  DOE  table.

And  we're  going  to  use  Join, going  to  join  it  with  our   curve data.

You  always  want  to  make  sure  that  you're matching  up  based  on  your  row  numbers,

so  the  right  curve  for  the  right  run  goes with  the  correct  factors  for  that  run.

And  I'm  going  to  select  all  the  columns

in  my  DOE  table  and  my  Functions from  my  Curve  Table.

A gain,  I  can  use  an  explicit name  here  when  I  create  this  table.

Now, I  have  my  table  ready  for  analysis.

That's  an  example  of  Join.

Let's  talk  about  Concatenate.

I  don't  use  concatenate  too  much for  my  DOE data  prep.

Concatenate…

You  use  that  when  you  want  to add  rows  to  a  data  table.

Then  DOE,

we  typically  have  all  of  our  rows, all  of  our  runs  in  our  data  tables.

We  don't  need  to  concatenate,

but  I  just  want  to  run  through this  example  real  quick.

Let's  say  I  have  my  data for  my  first  16  runs.

I  have  10  observations  per  run, so  I  have  160  rows.

Then  I  run  my  17th  run.

It looks like this.

That's 160.

Here's  my  17th  run  with the  10  observations  from  that  run.

Essentially,  what  I  want  to  do is  join  this  data  table,

or sorry,  concatenate.

I  want  to  add  these  rows at  the  bottom  of  this  data  table.

I  can  start  here, come  to  Concatenate.

We're  going  to  add  this.

With concatenate,  you  have  this option  to  append  to  first  table.

I'm  just  going  to  add  these  rows, append this  data  table.

Now,  we  have  170  rows  of  data.

That's  Concatenate.

I  want  to  end  up  here  with  Update.

Update  can  be  a  very  handy  tool

when  you're  populating your  DOE  table  with   curve data.

Here's  an  example of  the DOE  data table  I  created.

I  have  columns  here to  populate  my   curve data.

Here's  my   curve data.

Here's  my   DOE data table.

Essentially,  what  I  want  to  do  in  Update is  I  want  to  be  able  to  populate

my  blank  cell  with  the  information I  have  in  this  data  table.

I  can  do  that  by  matching  run  order,

and  then  JMP  will  automatically  match  up

the  columns  with  the  same  names and  update  this  data  table.

Let's  come  here in T ables,  Update.

Select  my  table  that  has  my   curve data.

Match  on  Run  Order, say  OK.

And  now,  this  data  table  is  updated.

Those  are  some  table  manipulations

you  can  do  to  get your  data  ready  for  analysis.

The  last  thing  I  want  to  talk  about is  importing  multiple  files.

Let's  say  that  your   curve data  gets  stored as  separate  files  for  each  batch.

I  have  this  example  here  of…

I  have  my   curve data   in  17  different  files,

and  they  happen  to  be  CSV  files.

I  want  to  be  able  to  import each  of  these  CSV  files

and  concatenate  them  together so  I  have  one  data  table.

You  can  easily  do  this  by  using

the  Import  Multiple  Files  function under  the  File  menu.

When you use Import  Multiple  File,

you  can  click  on  this  button  here

to  select  the  folder that  contains  all  of  those  files.

Here's  a  list  of  all  those  files.

Now,  the  file  name  itself  actually contains  my  batch  number,

and  this  is  data  that  I  actually want  to  pull  out  of  the  file  name.

I'm  going  to  add the  file  name  as  a  column.

We'll  import.

Here's  my   curve data with  time  and  size,

and  here's  my  file  name.

Now  I  can  come  up  to  the  Columns  menu

and  use  this  column  utility  to  convert my  text  to  multiple  columns.

I  just  have  to  put  in the  delimiter  I  want.

I'm  going  to  use  the  underscore that's  before  the  batch  number

and  the  dot  that's  after  the  batch  number, and  I  can  say  OK.

That  gave  me  three  columns:

the  curve,  the  batch  number, and  the  file  extension.

This  is  the  data  that  I  want.

I'm  just  going  to  delete these  other  columns  here.

And  now  I  have  all  of  my   curve data

for  all  my  17  runs  in  one  file with  the  batch  number.

That  is  what  I  wanted to  show  you  for  data  prep.

Now,  let's  talk  about setting  up  your  DOE  design.

There's  a  couple  of  tricks that  I  want  to  show  you.

In  your  DOE  Dialog, there's  two  things  to  think  about.

First  of  all,  we  want  to  make  sure we're  removing  this  default  response,

and  then  we're  going  to  talk  about how  to  define  the  functional  response

based  on  the  format  of  your   curve data.

Let's  go  ahead  and  launch  our  DOE  Dialog.

We're  going  to  come  up  here  to  DOE, go  to  Custom  Design.

Here's  our  DOE  dialog.

Now,  the  DOE  Dialog,  like  I  said, will  have  this  default  Y  response.

If  we  just  have a  functional  response  in  our  DOE,

we  don't  need  this  default  response, so  we  need  to  get  rid  of  this.

What  we  don't  want  to  do is  just  delete  the  name

because  that  response  is  still  there.

What  we  want  to  do  is  select that  default  response

and  actually  use  Remove  to  get  rid  of  it.

Then  we  want  to  add a  functional  response.

I'm  going  to  come  here and  add  a  functional  response.

When  we're  defining our  functional  response,

we  can  give  it  a  name.

We  can  say  the  number  of  measurements per  run  and  the  values.

Let's  go  ahead  and  do  this  for  our  DOE.

Our  responses   size…

This  is  what's  on  the  y- axis of  our  function.

Then  we  can  tell  the  DOE  platform what  our  X  values  look  like.

We  can  define  the  number  of  measurements

with  the  number  of  X  values and  what  those  X  values  are.

Let's  say  I'm  going  to  measure the  size  every  2  hours.

I'm  just  going  to  type  in  here every  2  hours  up  to  20  hours.

That  looks  good.

The  next  thing  I  need  to  do is  add  my  factors.

I  have  saved  my  factors and  my  factor  ranges  to  this  factor  table.

I'm  just  going  to  load  in  these  factors.

I  have  my  factors  up  here.

Next  thing  I  want  to  do is  specify  my  model.

I'm  going  to  choose a  response  surface  model

which  will  add  all  my  two- way interactions  and  all  my  quadratics.

Finally,  I  can  enter  in the  number  of  runs.

JMP  is  recommending a  default  number  of  21,

but  let's  say  I  only  have  enough time  and  resources  to  do  17.

I'll say 17.

I  will  ask  them  to  make  the  design  using all  of  those  inputs  that  I  entered.

It  just  takes  a  couple  of  seconds for  JMP  to  create  this  design  for  me.

Here's  my  design.

Here  are  my  17  runs  with  the  treatment I  want  to  apply  for  each  of  those  runs.

When  I'm  creating  my   DOE table,

I  always  want  to  use this  Make  Table  button.

I  always  like  to  include the  run  order  column

because  the  order  that  these  runs are  executed  is  very  important.

I'm  going  to  include  that  run  order  column and  make  our   DOE data  table.

Here's  our  DOE  data  table.

We  have  our  treatment.

We  have  a  place  for  R  to  enter in  results  for  our  function,

and  I  have  my  run  order  column here  at  the  end.

I  also  have  my  scripts

that  reflect  the  functional  DOE and  also  the  model  I  specified.

Whenever  I'm  adding  data, my   curve data,  like  I  said  before,

I  always  want  to  add  it  here to  my  DOE  data  table

because  it  contains  information about  the  functional  data  analysis

and  the  DOE  model  that  was  specified.

That's  a  quick  overview, but  I  want  to  give  you  some  tips  about

defining  the  functional  response based  on  the  format  of  your   curve data.

Let's  come  back  here.

We'll  go  back.

Let's  come  back  up  here  to  Responses.

The  way  that  you  populate this  information  here

will  define  how  your  DOE  data  table  looks.

I  want  to  give  you  a  couple  tips for  what  to  enter  here,

depending  on  what your   curve data  looks  like

because  we  want   the  data  prep  part to  be  as  easy  as  possible.

There's  a  couple  of  things  to  consider.

First  of  all,  is  your  curve data wide  or  tall?

We  talked  a  lot  about  this,  right?

Do  you  have  wide  data?

Or  do  you  have  tall  data?

Is  it  stacked  or  are  you  going  to have  rows  as  functions?

The  other  thing  to  consider  is  whether your  data  is  equally  spaced,

if  you  have  the  same  X  measurements,

or  whether  your  measurements are  asynchronous.

What  do  I  mean  here?

Well,  let  me  pull  up  a  couple  examples.

In  this  example  here, I  just  have  a  few  measurements  per  run.

I  have  10  measurements  per  run,

and  they  are  all  equally  spaced,

and  I  have  the  same measurements  for  each  run.

When  I  go  to  enter  in  this  information, there's  just  a  few  to  populate  here.

It's  not  that  difficult.

But  you  might  have  a  scenario where  you  have  asynchronous  data.

In  other  words,

you  have  different  measurements for  each  of  your  runs,

and  you  might  have  a  situation  where

you're  collecting  a  lot of  data  points,  maybe…

My  rule  of  thumb  is

if  you're  around  10,  less  than  20,  yeah, go  ahead  and  populate  your  values  here.

But  once  you  start  getting  up above  20,  certainly  hundreds,

that's  a  lot  of  information to  add  here  to  your  response  here.

The  other  thing  to  consider  is

if  your  data  will  be  manually  entered,

are  you  going  to  manually enter  the  responses

or  are  you  going  to  use  Join  or  Update?

Let's  run  through  some  scenarios.

Let's  say  you  have  rows as  functions.

You  have  a  few  measurements,

and  you're  going  to manually  enter  your  data.

Well,  if  you're  going  to  do  that, then  set  up  your  DOE  data table  like  this.

Or  set  up  your…

sorry,  your   functional  response  like  this.

This  is  what  your  DOE  data  table is  going  to  look  like.

You  can  manually  enter your  results  in  here,

and  then  you  can  use  this  script  here to  run  your  functional  data  analysis.

That's  the  first  scenario.

Let's  say  you  have  rows as  functions, you  have  a  few  measurements,

but  you  want  to  use  Update to  update  your  datas.

Again,  l et's  come  back   and  take  a  look  at  this  example  here.

In  this  example,

since  I  defined  the  name  of  my  response, I  get  time  in  here  with  my  column  header.

Let's  say  when  I  bring  in  my  data,

I  just  have  the  number  here in  my  column  header.

So  if  I  was  going  use  Update, these  column  names  do  not  match.

To  make  these  column  names  match,

what  I  want  to  do  is  come  back  here, remove  the  name,

and  then  when  I  create  my   DOE data  table, I  just  have  the  number.

And  then  I  can  use  Update to  update  this  data  table.

Let's  go  ahead  and  do  that.

Here's  my  data.

Let's  go  ahead  and  update with  my  current  data.

I'm  going  to  update  based  on matching  the  row  numbers,

and  now  my  data  is  in  here.

I  can  use  this  script  here  to  go  ahead

and  go  into  the  Functional  Data  Explorer to  start  analyzing  this  data.

So  that  scenario, let's  say  I  have  rows as  function.

Again,  I  have  wide  data, but  I  have  many  measurements.

I  have  many  more  than  10.

Let's  say  I  have  50.

Entering  the  50  values  in  here doesn't  make  a  whole  lot  of  sense.

What  I'm  going  to  do  is  I'm  going  to set  the  number  of  measurements  to  one,

and  I  can  just  set the  values  to  one  as  well.

When  I  go  to  make  my  DOE  data  table,

it  will  look  like  this.

I  will  get  my  run,  order.

I  will  get  my  factors, and  I'll  get  this  blank  column.

All  I  need  to  do  is  delete  that  column.

Here  are  my  50  measurements for  each  of  my  curves.

Again,  I'm  going  to  use  Join, like  I  showed  you  up  above.

I'm  going  to  match  based  on  run  order,

bring in  everything  from  my   DOE table, my  functions  from  my  curve  table.

I  can  give  it  an  explicit  name and  say  OK.

In  this  example  here, since  I've  used  Join,

I'm  essentially  ignoring  what  I  set  up  as the  details  around  my  functional  response.

This  script  here  is  not  going  to  work.

In  this  scenario, I  will  have  to  come  back  here,

go open  up  the  Functional  Data  Explorer, enter  my  supplementary  variables, my…

This  is  rows as  function.

Enter  in  my  supplementary  variables, my  run  ID,  and  my  curves.

When  I  go  to  do the  functional  DOE  analysis,

it  will  come  back  and  look  at the  model  here  that's  specified  here.

It's  generalized  regression  script,

so  it  will  remember  the  model  that specified  when  I  set  up  my  DOE.

That  is  that  example.

Let's  talk  about  stacked  format.

With  stacked  format,  we  typically  are going  to  be  adding  the  data  using  Join.

Again,  what  I  populate  here doesn't  really  matter,

just  as  long  as  I  have a  functional  response  entered  in  here.

Again,  I  get  this  same  data  table.

I  can  delete  out  the  response.

Oops, I added to it.

Delete column.

I  can  remove  that  response.

Now,  I  can  use  Join  to  bring  in my  stacked   curve data

by  matching  on  run  order,

bring in  everything  in  from  my  DOE  table and  my  function  data  from  my  curve  table.

Again,  for  this  example  here, running  this  script  is  not  going  to  work,

so I'm  going  to  have  to  manually  launch the  Functional  Data  Explorer,

bring  in  my  X,  my  Y,

my  supplementary  variable, and  my  run  order,

and  then  I  can  go  ahead  and  execute the  Functional  Data  Explorer

when  I  go  to  do  functional  DOE.

Again,  as  before,

it  will  look  at  the  model  that's  included in  generalized  regression  script

that  is  based  on  the  model  that  you specified  when  you  designed  your  DOE.

The  last  thing  I  want  to  mention real  quick  is  this FDE X Column Property

that  I  talked  about  before.

Let's  say  that…

this  is  a  scenario  where  I  want  to bring  in  data,  my   curve data,

where  I  have  rows as  functions.

So  I  have  rows as  functions, I  have  many  measurements.

I  want  to  add  the   curve data  using  Join, but  my  column  headings  contain  text.

In  this  case, I  have  the  units  of  measurement

for  each  of  my  X  values here  in  my  column  headers.

I  can  join  this  data  together,

bring in  my   curve data by  matching  on  run  order.

We then bring  in  all  my  data from  my  DOE  table,

bring in  all  of  my   curve data.

Let's  say  I  want

the  Functional  Data  Explorer  to  recognize the  number  in  my  column  header.

Well,  to  do  that, I  need  that   FDE X Column  Property.

But  when  I  go  in  here  to Column  Properties,

you're  not  going  to  find FDE X Column  Property  here.

What  I  can  do  is  I can use a script to define my FDE X Column Property.

Actually, it's going to be based on this.

What I can do is run  this  script, and  now  I  have  a  column  property  assigned

where  I  have  the  number  that  the  FDE will  recognize  as  your  X  value.

That  was  my  last  tip  or  trick.

Let's  just  do  a  quick  wrap- up, a  review  of  the  tips  and  tricks

starting  with  your  DOE  design.

You  always  want  to  remove that  default  Y  response

before  you  add  your  functional  response.

You're  going  to  define your  functional  response

based  on  the  format  of  your  curve data

because  you  want  to  make your  data  prep  as  easy  as  possible.

You  always  want  to  add your   curve data  to  the   DOE data  table

to  take  advantage  of  all, not  only  the  FDE  script,

but  the  model  script  that  is  created for  you  by  the  Custom  Design  Platform.

When  you're  preparing your  data  for  analysis,

when  you're  bringing  in  your  curve data, just  know  that

the  Functional  Data  Explorer accepts  different  formats,

stacked  data  and  rows  as  functions.

You  can  use  Stack,  Split,  Join, and  Update to  get  your  data  ready  for  analysis.

And  if  your   curve data is  stored  in  separate  files,

use  import  multiple  files.

I  just  want  to  acknowledge a  couple  of  people.

Ryan  Parker,  he  is  the  developer of  the  Functional  Data  Explorer.

I  want  to  acknowledge  him for  all  of  his  help  with  understanding

all  of  the  wonderful  things that  FDE  can  do.

I  also  want  to  thank  Chris  Gotwalt for  his  leadership,

also  for  some of  the  slides  that  I  used at  the  beginning  of  my  presentation.

With  that,  I  thank  you  very  much.

If  you  have  any  questions, I'd  love  to  hear  about  them  in  the  chat.

Thank  you.

Presenter