Applying Neural Fit from one data table to a different one

Question

I have two tables of data from different sources.  Both tables have the exact same set of column variables, but have different numbers of rows (one has 35 rows and the other has 150 rows).

I created a Neural Fit formula in JMP Standard for the larger dataset using 4 nodes and it resulted in a decent fit. I am now wanting to apply the exact formula created by the larger dataset to the smaller dataset.

The columns of data created by the large dataset look like this...

But when I copy the formulas into the smaller dataset then the results are broken...

I'm seem to be missing some sort of formatting or reference which is causing this error.  Any suggestions on how to port over the larger table's Neural Fit formula to the smaller table?

jthi · Answer

This will be easier option (if exclude scripting), use copy columns instead of copy column properties:

1. Save your formulas to table1

2. Select all the formula columns, right click and select copy columns

You will get script like this to your clipboard

View more... (Highlight to read)Paste Columns(
	{New Column("H1_1",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-33.6793185187104) + 0.116025148822519 * :Age + 0.257249853464313 * :BMI + 0.07709223029077 * :BP + -0.0111032508344931 *
				:Total Cholesterol + 0.00756992999069027 * :LDL + 0.0185393387261887 * :HDL + 0.109640405245873 * :TCH + 2.72330332145885 * :LTG
				+0.0040259120499412 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("H1_2",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-15.185081417965) + -0.116759523382996 * :Age + 0.209499698455059 * :BMI + 0.0495204569485005 * :BP + 0.000791158458320374 *
				:Total Cholesterol + -0.0252646330088984 * :LDL + -0.0435117437458085 * :HDL + -0.0337755173307676 * :TCH + 3.96793201576665 * :LTG
				 + -0.0289384701545602 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("H1_3",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-2.4417415178684) + 0.0413946703447229 * :Age + 0.279113099537121 * :BMI + -0.0663330287069535 * :BP + 0.00549686850152615 *
				:Total Cholesterol + 0.0095792805644581 * :LDL + 0.00158192809570205 * :HDL + -0.406068537243516 * :TCH + 0.274713939506156 * :LTG
				+ -0.00565455809670695 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("Predicted Y",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3) - Save Formulas"),
		Formula(137.891329398094 + 41.761323540748 * :H1_1 + 43.3118568935298 * :H1_2 + 29.861259819731 * :H1_3),
		Set Property("Predicting", {:Y, Creator("Neural"), Std Dev(50.1433686750535)}),
		Set Property("Expand Intermediates", 1),
		Set Selected
	)}
)

Paste Columns(
	{New Column("H1_1",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-33.6793185187104) + 0.116025148822519 * :Age + 0.257249853464313 * :BMI + 0.07709223029077 * :BP + -0.0111032508344931 *
				:Total Cholesterol + 0.00756992999069027 * :LDL + 0.0185393387261887 * :HDL + 0.109640405245873 * :TCH + 2.72330332145885 * :LTG
				+0.0040259120499412 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("H1_2",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-15.185081417965) + -0.116759523382996 * :Age + 0.209499698455059 * :BMI + 0.0495204569485005 * :BP + 0.000791158458320374 *
				:Total Cholesterol + -0.0252646330088984 * :LDL + -0.0435117437458085 * :HDL + -0.0337755173307676 * :TCH + 3.96793201576665 * :LTG
				 + -0.0289384701545602 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("H1_3",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3)"),
		Formula(
			TanH(
				0.5 * ((-2.4417415178684) + 0.0413946703447229 * :Age + 0.279113099537121 * :BMI + -0.0663330287069535 * :BP + 0.00549686850152615 *
				:Total Cholesterol + 0.0095792805644581 * :LDL + 0.00158192809570205 * :HDL + -0.406068537243516 * :TCH + 0.274713939506156 * :LTG
				+ -0.00565455809670695 * :Glucose)
			)
		),
		Set Property("Intermediate", 1),
		Set Selected
	), New Column("Predicted Y",
		Numeric,
		"Continuous",
		Format("Best", 12),
		Set Property("Notes", "Model NTanH(3) - Save Formulas"),
		Formula(137.891329398094 + 41.761323540748 * :H1_1 + 43.3118568935298 * :H1_2 + 29.861259819731 * :H1_3),
		Set Property("Predicting", {:Y, Creator("Neural"), Std Dev(50.1433686750535)}),
		Set Property("Expand Intermediates", 1),
		Set Selected
	)}
)
3. Go to your second table, right click and Paste columns

Now you should have the columns correctly in your new table

NathanFisk · Answer

1. I creating the Neural fit by using the "Save Formulas" option (so I can get each node's formula).

2. This creates five formula columns (4 for the nodes, 1 for the final prediction).

3. I create correlating columns in the smaller dataset manually for each of the five formula columns.
4. I go into the larger dataset, open each of the Neural formula column's "Properties" and then select "Formula" and copy the formula over into each of the five corresponding columns I made in the smaller dataset.
3. I also copy the column general properties over by just right clicking the column and copying the column properties.

jthi · Answer

How are you copying the columns over? Also do you need the separate formulas for hidden nodes as in have you tried the other formula saving options (Profile Formulas and Fast Formulas)?

Quick Links