step: Choose a model.

In Fuzzy-Logix/AdapteR: Converts R syntax to SQL to provide fast analytics using in-database C++ library DB Lytix.

Description

steps performs linear regression on FLTable objects. Choose a formula based model by p-values and R-Squared Values.

Usage

step(object, scope, ...)

Arguments

object	An object of class FLTable
scope	A symbolic description of model to be fitted. scope can be a list with upper and lower components or a formula. For a widetable, upper and lower should be formulas describing the range of models. If a formula is given instead of list it will be treated as upper. For a deeptable, upper and lower should be vectors with variable ids'.Provide empty list for deeptable if nothing is to be specified.
scale	currently not used.
direction	character.Must be one of backward, Fbackward,UFbackward,forward.
trace	if positive, information is printed out during the running of the steps.
catToDummy	Transform categorical variables to numerical values either using dummy variables or by using Empirical Logit. If the value is 1, transformation is done using dummy variables, else if the value is 0, transformation is done using Empirical Logit.
performNorm	0/1 indicating whether to perform standardization of data.
performVarReduc	0/1. If the value is 1, the stored procedure eliminates variables based on standard deviation and correlation.
makeDataSparse	If 0,Retains zeroes and NULL values from the input table. If 1, Removes zeroes and NULL. If 2,Removes zeroes but retains NULL values.
minStdDev	Minimum acceptable standard deviation for elimination of variables. Any variable that has a standard deviation below this threshold is eliminated. This parameter is only consequential if the parameter PerformVarReduc = 1. Must be >0.
maxCorrel	Maximum acceptable absolute correlation between a pair of columns for eliminating variables. If the absolute value of the correlation exceeds this threshold, one of the columns is not transformed. Again, this parameter is only consequential if the parameter PerformVarReduc = 1. Must be >0 and <=1.
classSpec	list describing the categorical dummy variables.
whereconditions	takes the where_clause as a string.
highestpAllow1	All the variables whose p-value exceed the value specified by HighestpAllow1 are dropped in one go. Typical value for HighestProbAllow1 could be 0.50. Must be >0 and < 1. Not applicable for forward.
highestpAllow2	Only one variable is dropped at a time till all the p-Values are below the HighestpAllow2. Typical value could be 0.10. Must be >0 and < 1. Not applicable for forward and backward.
stepWiseDecrease	The StepwiseDecrease is used to decrease the p-Value at each stage. In first step, all variables having pValue exceeding HighestpValue1 are dropped. Then the HighestpValue1 is reduced by StepwiseDecreasepValue and the process is repeated until all the variables have p-value less than HighestpValue2. Must be >0 and <1. Used only for UFbackward.

Value

step performs linear regression and replicates equivalent R output.

Constraints

The anova method is not yet available for FLLinRegr. Properties like print(fit$x),model,plot might take time as they have to fetch data

See Also

step for R reference implementation.

Examples

widetable  <- FLTable(getTestTableName("tblAbaloneWide"), "ObsID")
s <- step(widetable,
			 scope=list(lower=Rings~Height+Diameter),
	     direction = "UFbackward")
plot(s)
s$coefficients
s <- step(widetable,
			scope=list(lower=Rings~Height+Diameter,
 				   upper=Rings~Height+Diameter+Sex+Num_Length),
			direction = "UFbackward")
plot(s)
s$coefficients
s <- step(widetable,
		scope=list(lower=Rings~Num_Length),
			direction = "UFbackward",
		performNorm=1,performVarReduc=1,maxCorrel=0.6)
plot(s)
s$coefficients
s <- step(widetable,
			scope=list(upper=Rings~Height+Diameter+Sex+Num_Length+DummyCat),
 		direction = "Fbackward")
plot(s)
s$coefficients
s <- step(widetable,
			scope=Rings~Height+Diameter+Sex+Num_Length+DummyCat,
 		direction = "forward")
plot(s)
s$coefficients
s <- step(widetable,
			scope=Rings~Height+Diameter+Sex+Num_Length+DummyCat,
 		direction = "Fbackward")
plot(s)
s$coefficients
s <- step(widetable,
			scope=list(upper=Rings~Height+Diameter+Sex+Num_Length+DummyCat),
 		direction = "forward")
plot(s)
s$coefficients
deeptable <- FLTable(getTestTableName("myLinRegrSmall"),"ObsID","VarID","Num_Val")
s <- step(deeptable,
			scope=list(upper=c("-1","0","1")),
 		direction = "backward")
s <- step(deeptable,
			scope=list(upper=c("1","2"),lower=c("1")),
 		direction = "Fbackward")
s <- step(deeptable,
			scope=list(lower=c("2")),
 		direction = "UFbackward")
s <- step(deeptable,
			scope=list(),
 		direction = "forward")
deeptable1 <- FLTable(getTestTableName("tblLogRegr"),
					"ObsID","VarID","Num_Val",
                  whereconditions=c("ObsID < 7001","VarID<5"))
s <- step(deeptable1,
         scope=list(lower=c("2")),
         direction = "UFbackward",familytype = "logistic")
s <- step(deeptable1,
			scope=list(),
 		direction = "forward",familytype="logistic")
plot(s)
s <- step(deeptable1,
			scope=list(upper=c("-1","0","1","2","3")),
 		direction = "backward",
			familytype="multinomial",pRefLevel=1)
s <- step(deeptable1,
			scope=list(upper=c("1","2","3"),lower=c("2")),
 		direction = "Fbackward",familytype="multinomial",pRefLevel=1)
deeptable2 <- FLTable(getTestTableName("tblLogRegrMN10000"),
					"ObsID","VarID","Num_Val",
                  whereconditions=c("ObsID < 7001","VarID<5"))
s <- step(deeptable2,
         scope=list(lower=c("2")),
         direction = "UFbackward",familytype = "multinomial",pRefLevel=1)
summary(s)

Fuzzy-Logix/AdapteR documentation built on May 6, 2019, 5:07 p.m.