PivotalR between R & PostgreSQL-like Databases(for exp : Greenplum, hadoop access by hawq)
背景
PivotalR是R的一个包, 这个包提供了将R翻译成SQL语句的能力, 即对大数据进行挖掘的话. 用户将大数据存储在数据库中, 例如PostgreSQL , Greenplum.
用户在R中使用R的语法即可, 不需要直接访问数据库, 因为PivotalR 会帮你翻译成SQL语句, 并且返回结果给R.
这个过程不需要传输原始数据到R端, 所以可以完成R不能完成的任务(因为R是数据在内存中的运算, 如果数据量超过内存会有问题)
PivotalR还封装了MADlib, 里面包含了大量的机器学习的函数, 回归分析的函数等.
这个包的说明 :
PivotalR-package
An R font-end to PostgreSQL and Greenplum database, and wrapper
for in-database parallel and distributed machine learning open-source
library MADlib
Description
PivotalR is a package that enables users of R, the most popular open source statistical programming
language and environment to interact with the Pivotal (Greenplum) Database as well as Pivotal
HD/HAWQ for Big Data analytics. It does so by providing an interface to the operations on tables/views
in the database. These operations are almost the same as those of data.frame. Thus the
users of R do not need to learn SQL when they operate on the objects in the database. The latest
code is available at https://github.com/madlib-internal/PivotalR. A training video and a
quick-start guide are available at http://zimmeee.github.io/gp-r/#pivotalr.
Details
Package: PivotalR
Type: Package
Version: 0.1.17
Date: 2014-09-15
License: GPL (>= 2)
Depends: methods, DBI, RPostgreSQL
This package enables R users to easily develop, refine and deploy R scripts that leverage the parallelism
and scalability of the database as well as in-database analytics libraries to operate on big
data sets that would otherwise not fit in R memory - all this without having to learn SQL because
the package provides an interface that they are familiar with.
The package also provides a wrapper for MADlib. MADlib is an open-source library for scalable
in-database analytics. It provides data-parallel implementations of mathematical, statistical and
machine-learning algorithms for structured and unstructured data. The number of machine learning
algorithms that MADlib covers is quickly increasing.
As an R front-end to the PostgreSQL-like databases, this package minimizes the amount of data
transferred between the database and R. All the big data is stored in the database. The user enters
their familiar R syntax, and the package translates it into SQL queries and sends the SQL query into
database for parallel execution. The computation result, which is small (if it is as big as the original
data, what is the point of big data analytics?), is returned to R to the user.
On the other hand, this package also gives the usual SQL users the access of utilizing the powerful
analytics and graphics functionalities of R. Although the database itself has difficulty in plotting,
the result can be analyzed and presented beautifully with R.
This current version of PivotalR provides the core R infrastructure and data frame functions as well
as over 50 analytical functions in R that leverage in-database execution. These include
* Data Connectivity - db.connect, db.disconnect, db.Rquery
* Data Exploration - db.data.frame, subsets
* R language features - dim, names, min, max, nrow, ncol, summary etc
* Reorganization Functions - merge, by (group-by), samples
* Transformations - as.factor, null replacement
* Algorithms - linear regression and logistic regression wrappers for MADlib
Note
This package is differernt from PL/R, which is another way of using R with PostgreSQL-like
databases. PL/R enables the users to run R scripts from SQL. In the parallel Greenplum database,
one can use PL/R to implement parallel algorithms.
However, PL/R still requires non-trivial knowledge of SQL to use it effectively. It is mostly limited
to explicitly parallel jobs. And for the end user, it is still a SQL interface.
This package does not require any knowledge of SQL, and it works for both explicitly and implicitly
parallel jobs by employing the open-source MADlib library. It is much more scalable. And for the
end user, it is a pure R interface with the conventional R syntax.
Author(s)
Author: Predictive Analytics Team at Pivotal Inc. <user@madlib.net>, with contributions from
Data Scientist Team at Pivotal Inc.
Maintainer: Caleb Welton, Pivotal Inc. <cwelton@pivotal.io>
References
[1] MADlib website, http://madlib.net
[2] MADlib user docs, http://doc.madlib.net/master
[3] MADlib Wiki page, http://github.com/madlib/madlib/wiki
[4] MADlib contribution guide, https://github.com/madlib/madlib/wiki/Contribution-Guide
[5] MADlib on GitHub, https://github.com/madlib/madlib
See Also
madlib.lm Linear regression
madlib.glm Linear, logistic and multinomial logistic regressions
madlib.summary summary of a table in the database.
Examples
## Not run:
## get the help for the package
help("PivotalR-package")
## get help for a function
help(madlib.lm)
## create multiple connections to different databases
db.connect(port = 5433) # connection 1, use default values for the parameters
db.connect(dbname = "test", user = "qianh1", password = "", host =
"remote.machine.com", madlib = "madlib07", port = 5432) # connection 2
db.list() # list the info for all the connections
## list all tables/views that has "ornst" in the name
db.objects("ornst")
## list all tables/views
db.objects(conn.id = 1)
## create a table and the R object pointing to the table
## using the example data that comes with this package
delete("abalone", conn.id = cid)
x <- as.db.data.frame(abalone, "abalone")
## OR if the table already exists, you can create the wrapper directly
## x <- db.data.frame("abalone")
dim(x) # dimension of the data table
names(x) # column names of the data table
madlib.summary(x) # look at a summary for each column
lk(x, 20) # look at a sample of the data
## look at a sample sorted by id column
lookat(sort(x, decreasing = FALSE, x$id), 20)
lookat(sort(x, FALSE, NULL), 20) # look at a sample ordered randomly
## linear regression Examples --------
## fit one different model to each group of data with the same sex
fit1 <- madlib.lm(rings ~ . - id | sex, data = x)
fit1 # view the result
lookat(mean((x$rings - predict(fit1, x))^2)) # mean square error
## plot the predicted values v.s. the true values
ap <- x$rings # true values
ap$pred <- predict(fit1, x) # 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, "random"), 1000) # sort randomly
plot(random.sample) # plot a random sample
## fit a single model to all data treating sex as a categorical variable ---------
y <- x # make a copy, y is now a db.data.frame object
y$sex <- as.factor(y$sex) # y becomes a db.Rquery object now
fit2 <- madlib.lm(rings ~ . - id, data = y)
fit2 # view the result
lookat(mean((y$rings - predict(fit2, y))^2)) # mean square error
## logistic regression Examples --------
## fit one different model to each group of data with the same sex
fit3 <- madlib.glm(rings < 10 ~ . - id | sex, data = x, family = "binomial")
fit3 # view the result
## the percentage of correct prediction
lookat(mean((x$rings < 10) == predict(fit3, x)))
## fit a single model to all data treating sex as a categorical variable ----------
y <- x # make a copy, y is now a db.data.frame object
y$sex <- as.factor(y$sex) # y becomes a db.Rquery object now
fit4 <- madlib.glm(rings < 10 ~ . - id, data = y, family = "binomial")
fit4 # view the result
## the percentage of correct prediction
lookat(mean((y$rings < 10) == predict(fit4, y)))
## Group by Examples --------
## mean value of each column except the "id" column
lk(by(x[,-1], x$sex, mean))
## standard deviation of each column except the "id" column
lookat(by(x[,-1], x$sex, sd))
## Merge Examples --------
## create two objects with different rows and columns
key(x) <- "id"
y <- x[1:300, 1:6]
z <- x[201:400, c(1,2,4,5)]
## get 100 rows
m <- merge(y, z, by = c("id", "sex"))
lookat(m, 20)
## operator Examples --------
y <- x$length + x$height + 2.3
z <- x$length * x$height / 3
lk(y < z, 20)
## ------------------------------------------------------------------------
## Deal with NULL values
delete("null_data")
x <- as.db.data.frame(null.data, "null_data")
## OR if the table already exists, you can create the wrapper directly
## x <- db.data.frame("null_data")
dim(x)
names(x)
## ERROR, because of NULL values
fit <- madlib.lm(sf_mrtg_pct_assets ~ ., data = x)
## remove NULL values
y <- x # make a copy
for (i in 1:10) y <- y[!is.na(y[i]),]
dim(y)
fit <- madlib.lm(sf_mrtg_pct_assets ~ ., data = y)
fit
## Or we can replace all NULL values
x[is.na(x)] <- 45
## End(Not run)
安装,使用 :
> install.packages("PivotalR")
> library(PivotalR)
Loading required package: Matrix
Attaching package: ‘PivotalR’
The following objects are masked from ‘package:stats’:
sd, var
The following object is masked from ‘package:base’:
cbind
参考
1. http://blog.pivotal.io/data-science-pivotal/products/introducing-r-for-big-data-with-pivotalr
2. http://cran.r-project.org/web/packages/PivotalR/PivotalR.pdf
3. https://github.com/pivotalsoftware/PivotalR