PivotalR (version 0.1.18.5)

predict: Generate the db.Rquery object that can calculate the predictions

Description

Generate the db.Rquery object that can calculate the predictions for linear/logistic regressions. The actual result can be viewed using lk.

Usage

# S3 method for lm.madlib
predict(object, newdata, ...)
# S3 method for lm.madlib.grps
predict(object, newdata, ...)
# S3 method for logregr.madlib
predict(object, newdata, type = c("response",
                                  "prob"), ...)
# S3 method for logregr.madlib.grps
predict(object, newdata, type
= c("response", "prob"), ...)
# S3 method for glm.madlib
predict(object, newdata, type = c("response",
                                  "prob"), ...)
# S3 method for glm.madlib.grps
predict(object, newdata, type = c("response",
                                  "prob"), ...)

Arguments

object

The result of madlib.lm and madlib.glm.

newdata

A db.obj object, which contains the information about the real data in the database.

type

A string, default is "response". It produces the predicted results for the newdata. The alternative value is "prob", which is only used for binomial{logit} to compute the probabilities.

A string, default is "response", which produces the TRUE or FALSE prediction. If it is "prob", this function computes the probabilities for TRUE cases.

…

Extra parameters. Not implemented yet.

Value

A '>db.Rquery object, which contains the SQL query to compute the predictions.

See Also

madlib.lm linear regression

madlib.glm logistic regression

lk view the actual result

groups.lm.madlib, groups.lm.madlib.grps, groups.logregr.madlib, groups.logregr.madlib.grps extract grouping column information from the fitted model(s).

Examples

# NOT RUN {
# }
# NOT RUN {
<!-- %% @test .port Database port number -->

<!-- %% @test .dbname Database name -->
## set up the database connection
## Assume that .port is port number and .dbname is the database name
cid <- db.connect(port = .port, dbname = .dbname, verbose = FALSE)

## create db.table object pointing to a data table
delete("abalone", conn.id = cid)
x <- as.db.data.frame(abalone, "abalone", conn.id = cid, verbose = FALSE)

## Example 1 --------

fit <- madlib.lm(rings ~ . - sex - id, data = x)

fit

pred <- predict(fit, x) # prediction

content(pred)

ans <- x$rings # the actual value

lk((ans - pred)^2, 10) # squared error

lk(mean((ans - pred)^2)) # mean squared error

## Example 2 ---------

y <- x
y$sex <- as.factor(y$sex)
fit <- madlib.lm(rings ~ . - id, data = y)

lk(mean((y$rings - predict(fit, y))^2))

## Example 3 ---------

fit <- madlib.lm(rings ~ . - id | sex, data = x)

fit

pred <- predict(fit, x)

content(pred)

ans <- x$rings

lk(mean((ans - pred)^2))

## predictions for one group of data where sex = I
idx <- which(groups(fit)[["sex"]] == "I") # which sub-model
pred1 <- predict(fit[[idx]], x[x$sex == "I",]) # predict on part of data

## Example 3 --------

## plot the predicted values v.s. the true values
ap <- ans # true values
ap$pred <- pred # add a column which is the predicted values

## If the data set is very big, you do not want to load all the
## data points into R and plot. We can just plot a random sample.
random.sample <- lk(sort(ap, FALSE, NULL), 1000) # sort randomly

plot(random.sample)

## ------------------------------------------------------------
## GLM prediction

fit <- madlib.glm(rings ~ . - id | sex, data = x, family = poisson(log),
                  control = list(max.iter = 20))

p <- predict(f)

lk(p, 10)

db.disconnect(cid, verbose = FALSE)
# }