Problem extracting derivative of Spline fit in JSL

yvesprairie — Thu, 24 May 2018 21:01:07 GMT

HI everyone,

I am having a curious problem I cannot figure out. I have 2 variables (say Y and X) for which I want to extract the derivative of a smooth function linking the two. I can do it "manually" if my two variable are in a table. I simply fit a smooth spline, save the predicted formula, go in the formula editor and use the derivative option in the editor itself. Now I want to do the same but in an automated fashion because I will have dozens of such fitting to do so have reverted to using JSL. As far as I can tell, the only to get there in JSL is first to extract the spline coefficients as

splcoef=Spline Coef(X, Y, 1);

In the particular example that I am using this matrix would be:

[0.5 0.997009092716414 0.00052834644305232 0 0.0000661345992050551,

1 0.997281532762841 0.000577947392456002 0.0000992018988076818 0.0000439565382573196,

1.5 0.997600801501053 0.000710116694956763 0.000165136706193727

-0.0000523992394403174,

2 0.99799059412015 0.000835953971570102 0.0000865378470331725 -0.0000897493429406088,

2.5 0.998418986899826 0.00085517981139797 -0.0000480861673778753

-0.0000494911432825247,

3 0.99882836887077 0.000769975286558009 -0.000122322882301737 -0.0000204292892022263,

3.5 0.999180222132323 0.000632330437354731 -0.000152966816105107

-0.0000100612562586075,

4 0.999456887989942 0.000471817679055371 -0.000168058700493033 0.0000251846412034386,

4.5 0.999653930234497 0.000322647459465303 -0.000130281738687837

0.0000199230550287456,

5 0.999785173911436 0.000207308012048703 -0.000100397156144689 0.0000256287090160626,

5.5 0.999866932217051 0.000126132387666307 -0.0000619540926205567

0.0000114058197541642,

6 0.999915935615199 0.00007273265986134 -0.0000448453629892932 0.0000100640302564247,

7 0.999953886942327 0.0000132340246520343 -0.0000146532722199889

0.0000064471411305935,

7.5 0.999957646529239 0.0000034161082800262 -0.000004982560524089

0.0000026947675920361,

8 0.999958445789198 0.0000004546234498223 -0.0000009404091360308

0.0000012459427765851,

8.5 0.999958593741486 0.000000448671396435 0.0000009285050288488

-0.0000002521994682104,

9 0.999959018678507 0.0000011880268241599 0.0000005502058265328

-0.0000001515710315195,

9.5 0.999959731296997 0.0000016245543768191 0.0000003228492792533

-0.0000002152328528353,

10 0.999960597382399 0 0 0]

I can find the predicted values from the spline as:

predvalues=Spline Eval(X,splcoef);

to obtain the derivative, I tried to use:

dervalues=eval(derivative(spline eval(X, splcoef),X));

but this gives a unique value of zero (0).

However, if I replace the matrix reference splcoef by its actual contents in the spline eval statement then it works fine!!

dervalues=eval(derivative(spline eval(X,

[0.5 0.997009092716414 0.00052834644305232 0 0.0000661345992050551,

1 0.997281532762841 0.000577947392456002 0.0000992018988076818 0.0000439565382573196,

1.5 0.997600801501053 0.000710116694956763 0.000165136706193727

-0.0000523992394403174,

2 0.99799059412015 0.000835953971570102 0.0000865378470331725 -0.0000897493429406088,

2.5 0.998418986899826 0.00085517981139797 -0.0000480861673778753

-0.0000494911432825247,

3 0.99882836887077 0.000769975286558009 -0.000122322882301737 -0.0000204292892022263,

3.5 0.999180222132323 0.000632330437354731 -0.000152966816105107

-0.0000100612562586075,

4 0.999456887989942 0.000471817679055371 -0.000168058700493033 0.0000251846412034386,

4.5 0.999653930234497 0.000322647459465303 -0.000130281738687837

0.0000199230550287456,

5 0.999785173911436 0.000207308012048703 -0.000100397156144689 0.0000256287090160626,

5.5 0.999866932217051 0.000126132387666307 -0.0000619540926205567

0.0000114058197541642,

6 0.999915935615199 0.00007273265986134 -0.0000448453629892932 0.0000100640302564247,

7 0.999953886942327 0.0000132340246520343 -0.0000146532722199889

0.0000064471411305935,

7.5 0.999957646529239 0.0000034161082800262 -0.000004982560524089

0.0000026947675920361,

8 0.999958445789198 0.0000004546234498223 -0.0000009404091360308

0.0000012459427765851,

8.5 0.999958593741486 0.000000448671396435 0.0000009285050288488

-0.0000002521994682104,

9 0.999959018678507 0.0000011880268241599 0.0000005502058265328

-0.0000001515710315195,

9.5 0.999959731296997 0.0000016245543768191 0.0000003228492792533

-0.0000002152328528353,

10 0.999960597382399 0 0 0]

),X));

All I have done is replace the matrix name by its contents. This does not seem like normal behavior.

Any ideas how I get it to work??

Many thanks, Yves P.

Re: Problem extracting derivative of Spline fit in JSL

ms — Fri, 25 May 2018 00:27:21 GMT

Try the Eval( Evalexpr( ) ) cunstruct to force preevaluation of the matrix.

Names Default To Here(1);
X = 1 :: 100;
Y = Normal Distribution(X, 50, 25);
splcoef = Spline Coef(X, Y, 1);
predvalues = Spline Eval(X, Spline Coef(X, Y, 1));
dervalues = Eval(Eval Expr(Derivative(Spline Eval(X, Expr(splcoef)), X)));

// Plot to see if it works
dt = New Table("test", add rows(100));
dt << New Column("X", values(X)) << New Column("Y", values(predvalues)) <<
New Column("dY", values(dervalues));

dt << Graph Builder(
    Variables(X(:X), Y(:Y), Y(:dY)),
    Elements(Position(1, 1), Points(X, Y)),
    Elements(Position(1, 2), Points(X, Y))
);

Re: Problem extracting derivative of Spline fit in JSL

yvesprairie — Fri, 25 May 2018 00:33:53 GMT

Thank you so much, you just saved me hours of work!!! Sincerely, Yves

topic Problem extracting derivative of Spline fit in JSL in Discussions

Problem extracting derivative of Spline fit in JSL

Re: Problem extracting derivative of Spline fit in JSL

Re: Problem extracting derivative of Spline fit in JSL