HOW TO FIND MINIMA/MAXIMA IN JMP?

[#1]

Is there a way to find this minima(as shown in figure) and exclude data to its left.

[#2]

I'm not the right person to answer this, but it probably need a bit more information to understand what sort of answer fits:

Is the data really that smooth? Or is the data noisy?

Do you just need the first row that is bigger than the previous row?

Is there an equation generating the data?

[#3]

Hi Craige,

Thank you for the response. I have attached the data table( Minima_Jmp)  with X and Y columns for this graph.

Following are my responses:

1] Is the data really that smooth? Or is the data noisy?  

 Ans : Data is noisy - as shown in below 2 graphs

2] Do you just need the first row that is bigger than the previous row? 

Ans: This question isnt very clear to me - I would find the first minima point and exclude all the data to its left

3] Is there an equation generating the data? 

And: No there isnt an equation for this data - Maybe we can fit an equation to describe the curve but that only be an approximation 

[#4]

Your original question appears to be about a spline, fit to the noisy data, and finding the first minimum of that spline. For the example data, this seems to work:

dt = Open( "Z:/Minima_JMP.jmp" );
dt << sort( by( x ), replacetable( 1 ) );
obj = dt << Bivariate(
	Y( :Y ),
	X( :X ),
	Fit Spline( 26.8534445, {Line Color( {66, 112, 221} )} ),
	Automatic Recalc( 1 ),
	SendToReport( Dispatch( {}, "Bivar Plot", FrameBox, {Frame Size( 1653, 240 )} ) )
);
obj << (Curve["Smoothing Spline Fit, lambda=26.8534445"] << Save Predicteds);

For( irow = 2, irow < N Rows( dt ), irow += 1,
	If( dt:Spline Predictor for Y[irow] > dt:Spline Predictor for Y[irow - 1],
		Write( Eval Insert( "the inflection upward trend begins around row ^irow^" ) );
		// choose one of these...
		dt << selectrows( 1 :: irow );
//		dt<<deleterows(1::irow);
		Break();
	)
);
If( irow >= N Rows( dt ),
	Write( "not found?" )
);

The selected points in the graph are also selected rows in the data table; press delete in the table. Or, choose the <<deleterows statement.

In the log: the inflection upward trend begins around row 231

That is after sorting the table by ascending X.

You'll get different answers by adjusting the spline's tension (lambda=...). Make sure to delete the predicted column before rerunning the script, or close the table without saving.

This idea will break if the actual shape of the data doesn't match your example; if it is already trending up at the beginning, etc.

This may not be the correct way to do what you need; I think you'll have to explain what sort of process the data comes from and why you want to find this point in the data in order to get other people to add some ideas.

[#5]

Hi Craige,

This looks wonderful. I will try it on other data tables and check. 

Is there a way we can find a suitable lambda value if I have to automate this analysis for multiple data tables of similar data?

Thank you.

[#6]

I don't know. I think it is part science, part art, especially with splines. I think you'll want to add some sanity checks to the code if you try to automate it without visually inspecting the results. If it is possible for your process to produce an extra oscillation at the beginning, this will clearly fail. Or more data, or less data, or differently dense/sparse over the range, etc. might change the results.

There are people that might suggest other models that are specific to processes or industries. They would need more information about the data.