Vectors and arrays

Beside functions and environments, most of the objects an R user is interacting with are vector-like. For example, this means that any scalar is in fact a vector of length one.

The class Vector has a constructor:

>>> x = robjects.Vector(3)
class rpy2.robjects.Vector(o)[source]

Bases: rpy2.robjects.robject.RObjectMixin

Vector(seq) -> Vector.

The parameter ‘seq’ can be an instance inheriting from rinterface.SexpVector, or an arbitrary Python object. In the later case, a conversion will be attempted using conversion.py2rpy().

R vector-like object. Items can be accessed with:

  • the method “__getitem__” (“[” operator)

  • the delegators rx or rx2


iterator on names and values


Names for the items in the vector.

sample(n, replace: bool = False, probabilities=None)[source]

Draw a random sample of size n from the vector.

If ‘replace’ is True, the sampling is done with replacement. The optional argument ‘probabilities’ can indicate sampling probabilities.

Creating vectors

Creating vectors can be achieved either from R or from Python.

When the vectors are created from R, one should not worry much as they will be exposed as they should by rpy2.robjects.

When one wants to create a vector from Python, either the class Vector or the convenience classes IntVector, FloatVector, BoolVector, StrVector can be used.

class rpy2.robjects.vectors.BoolVector(obj)[source]

Bases: rpy2.robjects.vectors.Vector, rpy2.rinterface.BoolSexpVector

Vector of boolean (logical) elements BoolVector(seq) -> BoolVector.

The parameter ‘seq’ can be an instance inheriting from rinterface.SexpVector, or an arbitrary Python sequence. In the later case, all elements in the sequence should be either booleans, or have a bool() representation.

class rpy2.robjects.vectors.IntVector(obj)[source]

Bases: rpy2.robjects.vectors.Vector, rpy2.rinterface.IntSexpVector

Vector of integer elements IntVector(seq) -> IntVector.

The parameter ‘seq’ can be an instance inheriting from rinterface.SexpVector, or an arbitrary Python sequence. In the later case, all elements in the sequence should be either integers, or have an int() representation.


Like the R function tabulate, count the number of times integer values are found

class rpy2.robjects.vectors.FloatVector(obj)[source]

Bases: rpy2.robjects.vectors.Vector, rpy2.rinterface.FloatSexpVector

Vector of float (double) elements

FloatVector(seq) -> FloatVector.

The parameter ‘seq’ can be an instance inheriting from rinterface.SexpVector, or an arbitrary Python sequence. In the later case, all elements in the sequence should be either float, or have a float() representation.

class rpy2.robjects.vectors.StrVector(obj)[source]

Bases: rpy2.robjects.vectors.Vector, rpy2.rinterface_lib.sexp.StrSexpVector

Vector of string elements

StrVector(seq) -> StrVector.

The parameter ‘seq’ can be an instance inheriting from rinterface.SexpVector, or an arbitrary Python sequence. In the later case, all elements in the sequence should be either strings, or have a str() representation.

factor() → FactorVector[source]

Construct a factor vector from a vector of strings.

class rpy2.robjects.vectors.ListVector(obj)[source]

Bases: rpy2.robjects.vectors.Vector, rpy2.rinterface.ListSexpVector

R list (vector of arbitray elements)

ListVector(itemable) -> ListVector.

The parameter ‘itemable’ can be:

  • an object with a method items(), such for example a dict, a rpy2.rlike.container.TaggedList, an rpy2.rinterface.SexpVector of type VECSXP.

  • an iterable of (name, value) tuples

static from_length(length)[source]

Create a list of given length

Sequences of date or time points can be stored in POSIXlt or POSIXct objects. Both can be created from Python sequences of time.struct_time objects or from R objects.

class rpy2.robjects.vectors.POSIXlt(obj)[source]

Bases: rpy2.robjects.vectors.POSIXt, rpy2.robjects.vectors.ListVector

Representation of dates with a 11-component structure (similar to Python’s time.struct_time).

POSIXlt(seq) -> POSIXlt.

The constructor accepts either an R vector or a sequence (an object with the Python sequence interface) of time.struct_time objects.

class rpy2.robjects.vectors.POSIXct(obj)[source]

Bases: rpy2.robjects.vectors.POSIXt, rpy2.robjects.vectors.FloatVector

Representation of dates as seconds since Epoch. This form is preferred to POSIXlt for inclusion in a DataFrame.

POSIXlt(seq) -> POSIXlt.

The constructor accepts either an R vector floats or a sequence (an object with the Python sequence interface) of time.struct_time objects.

static sexp_from_datetime(seq)[source]

return a POSIXct vector from a sequence of datetime.datetime elements.

New in version 2.2.0: Vectors for date or time points


R’s factors are somewhat peculiar: they aim at representing a memory-efficient vector of labels, and in order to achieve it are implemented as vectors of integers to which are associated a (presumably shorter) vector of labels. Each integer represents the position of the label in the associated vector of labels.

For example, the following vector of labels







will become











>>> sv = ro.StrVector('ababbc')
>>> fac = ro.FactorVector(sv)
>>> print(fac)
[1] a b a b b c
Levels: a b c
>>> tuple(fac)
(1, 2, 1, 2, 2, 3)
>>> tuple(fac.levels)
('a', 'b', 'c')

Since a FactorVector is an IntVector with attached metadata (the levels), getting items Python-style was not changed from what happens when gettings items from a IntVector. A consequence to that is that information about the levels is then lost.

>>> item_i = 0
>>> fac[item_i]

Getting the level corresponding to an item requires using the levels,:

>>> fac.levels[fac[item_i] - 1]


Do not forget to subtract one to the value in the FactorVector. Indexing in Python starts at zero while indexing R starts at one, and recovering the level for an item requires an adjustment between the two.

When extracting elements from a FactorVector a sensible default might be to use R-style extracting (see Extracting items), as it preserves the integer/string duality.

class rpy2.robjects.vectors.FactorVector(obj)[source]

Bases: rpy2.robjects.vectors.IntVector

Vector of ‘factors’.


levels = rinterface.MissingArg, labels = rinterface.MissingArg, exclude = rinterface.MissingArg, ordered = rinterface.MissingArg) -> FactorVector

obj: StrVector or StrSexpVector levels: StrVector or StrSexpVector labels: StrVector or StrSexpVector (of same length as levels) exclude: StrVector or StrSexpVector ordered: boolean


are the levels in the factor ordered ?


Iterate the over the labels, that is iterate over the items returning associated label for each item


number of levels

Extracting items

Extracting elements of sequence/vector can become a thorny issue as Python and R differ on a number of points (index numbers starting at zero / starting at one, negative index number meaning index from the end / everything except, names cannot / can be used for subsettting).

In order to solve this, the Python way and the R way were made available through two different routes.

Extracting, Python-style

The python __getitem__() method behaves like a Python user would expect it for a vector (and indexing starts at zero).

>>> x = robjects.r.seq(1, 5)
>>> tuple(x)
(1, 2, 3, 4, 5)
>>> x.names = robjects.StrVector('abcde')
>>> print(x)
a b c d e
1 2 3 4 5
>>> x[0]
>>> x[4]
>>> x[-1]

Extracting, R-style

Access to R-style extracting/subsetting is granted though the two delegators rx and rx2, representing the R functions [ and [[ respectively (See the note below about [[, and $).

In short, R-style extracting has the following characteristics:

  • indexing starts at one (while Python indexing starts at zero).

  • the argument to subset on can be a vector of

    • integers (negative integers meaning exlusion of the elements)

    • booleans

    • strings (whenever the vector has names for its elements)

>>> print(x.rx(1))
[1] 1
>>> print(x.rx('a'))

R can extract several elements at once:

>>> i = robjects.IntVector((1, 3))
>>> print(x.rx(i))
[1] 1 3
>>> b = robjects.BoolVector((False, True, False, True, True))
>>> print(x.rx(b))
[1] 2 4 5

When a boolean extract vector is of smaller length than the vector, is expanded as necessary (this is know in R as the recycling rule):

>>> print(x.rx(True))
>>> b = robjects.BoolVector((False, True))
>>> print(x.rx(b))
[1] 2 4

In R, negative indices are understood as element exclusion.

>>> print(x.rx(-1))
>>> i = robjects.IntVector((-1, -3))
>>> print(x.rx(i))
[1] 2 4 5

That last example could also be written:

>>> i = - robjects.IntVector((1, 3)).ro
>>> print(x.rx(i))
[1] 2 4 5

This extraction system is quite expressive, as it allows a very simple writting of very common tasks in data analysis such as reordering and random sampling.

>>> from rpy2.robjects.packages import importr
>>> base = importr('base')
>>> x = robjects.IntVector((5,3,2,1,4))
>>> o_i = base.order(x)
>>> print(x.rx(o_i))
[1] 1 2 3 4 5
>>> rnd_i = base.sample(x)
>>> x_resampled = x.rx(o_i)

R operators are vector operations, with the operator applied to each element in the vector. This can be used to build extraction indexes.

>>> i = > 3 # extract values > 3
>>> i = ( >= 2 ).ro & ( <= 4) # extract values between 2 and 4

(More on R operators in Section Operators).

R/S also have particularities, in which some see consistency issues. For example although the indexing starts at 1, indexing on 0 does not return an index out of bounds error but a vector of length 0:

>>> print(x.rx(0))


What about the R operator $ ? In R, elements of a list can be extracted with [, or if only one element is wanted [[ or $ (with [[ able to extract on index, that is position in the list, or on name while $ can only extract on name).

In R, the 3 ways to extract one element out of a list are:

> l <- list(a = 1:3, b = 4:6)
> l[[1]]
[1] 1 2 3
> l[["a"]]
[1] 1 2 3
> l$a
[1] 1 2 3

With rpy2, it is looking like:

>>> elt = l.rx2(1) # This is the R `[[`, so one-offset indexing
>>> elt = l.rx2('a')

Assigning items

Assigning, Python-style

Since vectors are exposed as Python mutable sequences, the assignment works as for regular Python lists.

>>> x = robjects.IntVector((1,2,3))
>>> print(x)
[1] 1 2 3
>>> x[0] = 9
>>> print(x)
[1] 9 2 3

In R vectors can be named, that is elements of the vector have a name. This is notably the case for R lists. Assigning based on names can be made easily by using the method Vector.index(), as shown below.

>>> x = robjects.ListVector({'a': 1, 'b': 2, 'c': 3})
>>> x[x.names.index('b')] = 9


Vector.index() has a complexity linear in the length of the vector’s length; this should be remembered if performance issues are met.

Assigning, R-style

Differences between the two languages require few adaptations, and in appearance complexify a little the task. Should other Python-based systems for the representation of (mostly numerical) data structure, such a numpy be preferred, one will be encouraged to expose our rpy2 R objects through those structures.

The attributes rx and rx2 used previously can again be used:

>>> x = robjects.IntVector(range(1, 4))
>>> print(x)
[1] 1 2 3
>>> x.rx[1] = 9
>>> print(x)
[1] 9 2 3

For the sake of complete compatibility with R, arguments can be named (and passed as a dict or rpy2.rlike.container.TaggedList).

>>> x = robjects.ListVector({'a': 1, 'b': 2, 'c': 3})
>>> x.rx2[{'i': x.names.index('b')}] = 9

Missing values

Anyone with experience in the analysis of real data knows that some of the data might be missing. In S/Splus/R special NA values can be used in a data vector to indicate that fact, and rpy2.robjects makes aliases for those available as data objects NA_Logical, NA_Real, NA_Integer, NA_Character, NA_Complex.

>>> x = robjects.IntVector(range(3))
>>> x[0] = robjects.NA_Integer
>>> print(x)
[1] NA  1  2

The translation of NA types is done at the item level, returning a pointer to the corresponding NA singleton class.

>>> x[0] is robjects.NA_Integer
>>> x[0] == robjects.NA_Integer
>>> [y for y in x if y is not robjects.NA_Integer]
[1, 2]


NA_Logical is the alias for R’s NA.


The NA objects are imported from the corresponding rpy2.rinterface objects.


Mathematical operations on two vectors: the following operations are performed element-wise in R, recycling the shortest vector if, and as much as, necessary.

To expose that to Python, a delegating attribute ro is provided for vector-like objects.












** or ^















>>> x = robjects.r.seq(1, 10)
>>> print( + 1)


In Python, using the operator + on two sequences concatenates them and this behavior has been conserved:

>>> print(x + 1)
[1]  1  2  3  4  5  6  7  8  9 10  1


The boolean operator not cannot be redefined in Python (at least up to version 2.5), and its behavior could not be made to mimic R’s behavior


R vectors can have a name given to all or some of the elements. The property names can be used to get, or set, those names.

>>> x = robjects.r.seq(1, 5)
>>> x.names = robjects.StrVector('abcde')
>>> x.names[0]
>>> x.names[0] = 'z'
>>> tuple(x.names)
('z', 'b', 'c', 'd', 'e')


In R, arrays are simply vectors with a dimension attribute. That fact was reflected in the class hierarchy with robjects.Array inheriting from robjects.Vector.

class rpy2.robjects.vectors.Array(obj)[source]

Bases: rpy2.robjects.vectors.Vector

An R array


names associated with the dimension.


names associated with the dimension.


A Matrix is a special case of Array. As with arrays, one must remember that this is just a vector with dimension attributes (number of rows, number of columns).

>>> m = robjects.r.matrix(robjects.IntVector(range(10)), nrow=5)
>>> print(m)
     [,1] [,2]
[1,]    0    5
[2,]    1    6
[3,]    2    7
[4,]    3    8
[5,]    4    9


In R, matrices are column-major ordered, although the constructor matrix() accepts a boolean argument byrow that, when true, will build the matrix as if row-major ordered.

Computing on matrices

Regular operators work element-wise on the underlying vector.

>>> m = robjects.r.matrix(robjects.IntVector(range(4)), nrow=2)
>>> print( + 1)
     [,1] [,2]
[1,]    1    3
[2,]    2    4

For more on operators, see Operators.

Matrix multiplication is available as, transposition as Matrix.transpose(). Common operations such as cross-product, eigen values computation , and singular value decomposition are also available through method with explicit names.

>>> print( m.crossprod(m) )
     [,1] [,2]
[1,]    1    3
[2,]    3   13
>>> print( m.transpose().dot(m) )
     [,1] [,2]
[1,]    1    3
[2,]    3   13
class rpy2.robjects.vectors.Matrix(obj)[source]

Bases: rpy2.robjects.vectors.Array

An R matrix


Column names


crossproduct X’.Y


Matrix multiplication


Eigen values


Number of columns


Number of rows


Row names

svd(nu=None, nv=None, linpack=False)[source]

SVD decomposition. If nu is None, it is given the default value min(tuple(self.dim)). If nv is None, it is given the default value min(tuple(self.dim)).


crossproduct X.Y’


transpose the matrix


Extracting can still be performed Python-style or R-style.

>>> m = ro.r.matrix(ro.IntVector(range(2, 8)), nrow=3)
>>> print(m)
     [,1] [,2]
[1,]    2    5
[2,]    3    6
[3,]    4    7
>>> m[0]
>>> m[5]
>>> print(m.rx(1))
[1] 2
>>> print(m.rx(6))
[1] 7

Matrixes are two-dimensional arrays, and elements can be extracted according to two indexes:

>>> print(m.rx(1, 1))
[1] 2
>>> print(m.rx(3, 2))
[1] 7

Extracting a whole row, or column can be achieved by replacing an index number by True or False

Extract the first column:

>>> print(m.rx(True, 1))

Extract the second row:

>>> print(m.rx(2, True))


Data frames are a common way in R to represent the data to analyze.

A data frame can be thought of as a tabular representation of data, with one variable per column, and one data point per row. Each column is an R vector, which implies one type for all elements in one given column, and which allows for possibly different types across different columns.

If we consider for example tre data about pharmacokinetics of theophylline in different subjects, the data table could look like this:


























Such data representation shares similarities with a table in a relational database: the structure between the variables, or columns, is given by other column. In the example above, the grouping of measures by subject is given by the column Subject.

In rpy2.robjects, DataFrame represents the R class data.frame.

Creating objects

Creating a DataFrame can be done by:

  • Using the constructor for the class

  • Create the data.frame through R

  • Read data from a file using the instance method from_csvfile()

The DataFrame constructor accepts either an rinterface.SexpVector (with typeof equal to VECSXP, that is, an R list) or any Python object implementing the method items() (for example dict or rpy2.rlike.container.OrdDict).

Empty data.frame:

>>> dataf = robjects.DataFrame({})

data.frame with 2 two columns (not that the order of the columns in the resulting DataFrame can be different from the order in which they are declared):

>>> d = {'a': robjects.IntVector((1,2,3)), 'b': robjects.IntVector((4,5,6))}
>>> dataf = robject.DataFrame(d)

To create a DataFrame and be certain of the clumn order order, an ordered dictionary can be used:

>>> import rpy2.rlike.container as rlc
>>> od = rlc.OrdDict([('value', robjects.IntVector((1,2,3))),
                      ('letter', robjects.StrVector(('x', 'y', 'z')))])
>>> dataf = robjects.DataFrame(od)
>>> print(dataf.colnames)
[1] "letter" "value"

Creating the data.frame in R can otherwise be achieved in numerous ways, as many R functions do return a data.frame, such as the function data.frame().


When creating a DataFrame, vectors of strings are automatically converted by R into instances of class Factor. This behavior can be prevented by wrapping the call into the R base function I.

from rpy2.robjects.vectors import DataFrame, StrVector
from rpy2.robjects.packages import importr
base = importr('base')
dataf = DataFrame({'string': base.I(StrVector('abbab')),
                   'factor': StrVector('abbab')})

Here the DataFrame dataf now has two columns, one as a Factor, the other one as a StrVector

>>> dataf.rx2('string')
<StrVector - Python:0x95fe5ec / R:0x9646ea0>
>>> dataf.rx2('factor')
<FactorVector - Python:0x95fe86c / R:0x9028138>

Extracting elements

Here again, Python’s __getitem__() will work as a Python programmer will expect it to:

>>> len(dataf)
>>> dataf[0]
<Vector - Python:0x8a58c2c / R:0x8e7dd08>

The DataFrame is composed of columns, with each column being possibly of a different type:

>>> [column.rclass[0] for column in dataf]
['factor', 'integer']

Using R-style access to elements is a little richer, with the rx2 accessor taking more importance than earlier.

Like with Python’s __getitem__() above, extracting on one index selects columns:

>>> dataf.rx(1)
<DataFrame - Python:0x8a584ac / R:0x95a6fb8>
>>> print(dataf.rx(1))
1      x
2      y
3      z

Note that the result is itself of class DataFrame. To get the column as a vector, use rx2:

>>> dataf.rx2(1)
<Vector - Python:0x8a4bfcc / R:0x8e7dd08>
>>> print(dataf.rx2(1))
[1] x y z
Levels: x y z

Since data frames are table-like structure, they can be thought of as two-dimensional arrays and can therefore be extracted on two indices.

>>> subdataf = dataf.rx(1, True)
>>> print(subdataf)
  letter value
1      x     1
>>> rows_i <- robjects.IntVector((1,3))
>>> subdataf = dataf.rx(rows_i, True)
>>> print(subdataf)
  letter value
1      x     1
3      z     3

That last example is extremely common in R. A vector of indices, here rows_i, is used to take a subset of the DataFrame.

Python docstrings

class rpy2.robjects.vectors.DataFrame(tlist)[source]

Bases: rpy2.robjects.vectors.ListVector

R ‘data.frame’.

cbind(*args, **kwargs)[source]

bind objects as supplementary columns

static from_csvfile(path, header=True, sep=',', quote='"', dec='.', row_names=<rpy2.rinterface._MissingArgType object> [RTYPES.SYMSXP], col_names=<rpy2.rinterface._MissingArgType object> [RTYPES.SYMSXP], fill=True, comment_char='', na_strings=[], as_is=False)[source]

Create an instance from data in a .csv file.

path : string with a path header : boolean (heading line with column names or not) sep : separator character quote : quote character row_names : column name, or column index for column names

(warning: indexing starts at one in R)

fill : boolean (fill the lines when less entries than columns) comment_char : comment character na_strings : a list of strings which are interpreted to be NA values as_is : boolean (keep the columns of strings as such, or turn

them into factors)

head(*args, **kwargs)[source]

Call the R generic ‘head()’.


iterator across columns


iterator across rows


Number of columns. :rtype: integer


Number of rows. :rtype: integer

rbind(*args, **kwargs)[source]

bind objects as supplementary rows


Row names

to_csvfile(path, quote=True, sep=', ', eol='\n', na='NA', dec='.', row_names=True, col_names=True, qmethod='escape', append=False)[source]

Save the data into a .csv file.

path : string with a path quote : quote character sep : separator character eol : end-of-line character(s) na : string for missing values dec : string for decimal separator row_names : boolean (save row names, or not) col_names : boolean (save column names, or not) comment_char : method to ‘escape’ special characters append : boolean (append if the file in the path is

already existing, or not)