from functools import partial
from rpy2.ipython import html
html.html_rdataframe=partial(html.html_rdataframe, table_class="docutils")

dplyr in Python¶

We need 2 things for this:

1- A data frame (using one of R's demo datasets).

from rpy2.robjects.packages import importr, data
datasets = importr('datasets')
mtcars_env = data(datasets).fetch('mtcars')
mtcars = mtcars_env['mtcars']

In addition to that, and because this tutorial is in a notebook, we initialize HTML rendering for R objects (pretty display of R data frames).

import rpy2.ipython.html
rpy2.ipython.html.init_printing()

2- dplyr

from rpy2.robjects.lib.dplyr import DataFrame

/home/laurent/Desktop/software/python/py36_env/lib/python3.6/site-packages/rpy2-3.0.0-py3.6.egg/rpy2/robjects/lib/dplyr.py:24: UserWarning: This was designed againt dplyr version 0.8.0.1 but you have 0.7.7

With this we have the choice of chaining (D3-style)

dataf = (DataFrame(mtcars).
         filter('gear>3').
         mutate(powertoweight='hp*36/wt').
         group_by('gear').
         summarize(mean_ptw='mean(powertoweight)'))

dataf

or with pipes (magrittr style).

# currently no longer working
from rpy2.robjects.lib.dplyr import (filter,
                                     mutate,
                                     group_by,
                                     summarize)

if False:
    dataf = (DataFrame(mtcars) >>
             filter('gear>3') >>
             mutate(powertoweight='hp*36/wt') >>
             group_by('gear') >>
             summarize(mean_ptw='mean(powertoweight)'))

    dataf

The strings passed to the dplyr function are evaluated as expression, just like this is happening when using dplyr in R. This means that when writing mean(powertoweight) the R function mean() is used.

Using a Python function is not too difficult though. We can just call Python back from R. To achieve this we simply use the decorator rternalize.

# Define a python function, and make
# it a function R can use through `rternalize`
from rpy2.rinterface import rternalize
@rternalize
def mean_np(x):
    import statistics
    return statistics.mean(x)

# Bind that function to a symbol in R's
# global environment
from rpy2.robjects import globalenv
globalenv['mean_np'] = mean_np

# Write a dplyr chain of operations,
# using our Python function `mean_np`
dataf = (DataFrame(mtcars).
         filter('gear>3').
         mutate(powertoweight='hp*36/wt').
         group_by('gear').
         summarize(mean_ptw='mean(powertoweight)',
                   mean_np_ptw='mean_np(powertoweight)'))

dataf

It is also possible to carry this out without having to place the custom function in R's global environment.

del(globalenv['mean_np'])

from rpy2.robjects.lib.dplyr import StringInEnv
from rpy2.robjects import Environment
my_env = Environment()
my_env['mean_np'] = mean_np

dataf = (DataFrame(mtcars).
         filter('gear>3').
         mutate(powertoweight='hp*36/wt').
         group_by('gear').
         summarize(mean_ptw='mean(powertoweight)',
                   mean_np_ptw=StringInEnv('mean_np(powertoweight)',
                                           my_env)))

dataf

note: rpy2's interface to dplyr is implementing a fix to the (non-?)issue 1323 (https://github.com/hadley/dplyr/issues/1323)

The seamless translation of transformations to SQL whenever the data are in a table can be used directly. Since we are lifting the original implementation of dplyr, it just works.

from rpy2.robjects.lib.dplyr import dplyr
# in-memory SQLite database broken in dplyr's src_sqlite
# db = dplyr.src_sqlite(":memory:")
import tempfile
with tempfile.NamedTemporaryFile() as db_fh:
    db = dplyr.src_sqlite(db_fh.name)
    # copy the table to that database
    dataf_db = DataFrame(mtcars).copy_to(db, name="mtcars")
    res = (dataf_db.
           filter('gear>3').
           mutate(powertoweight='hp*36/wt').
           group_by('gear').
           summarize(mean_ptw='mean(powertoweight)'))
    print(res)
#

# Source:   lazy query [?? x 2]
# Database: sqlite 3.22.0 [/tmp/tmpojj60037]
   gear mean_ptw
  <dbl>    <dbl>
1     4    1237.
2     5    2574.

Since we are manipulating R objects, anything available to R is also available to us. If we want to see the SQL code generated that's:

silent = dplyr.show_query(res)

R[write to console]: <SQL>
SELECT `gear`, AVG(`powertoweight`) AS `mean_ptw`
FROM (SELECT `mpg`, `cyl`, `disp`, `hp`, `drat`, `wt`, `qsec`, `vs`, `am`, `gear`, `carb`, `hp` * 36.0 / `wt` AS `powertoweight`
FROM (SELECT *
FROM `mtcars`
WHERE (`gear` > 3.0)))
GROUP BY `gear`

The conversion rules in rpy2 make the above easily applicable to pandas data frames, completing the "lexical loan" of the dplyr vocabulary from R.

from rpy2.robjects import pandas2ri
from rpy2.robjects import default_converter
from rpy2.robjects.conversion import localconverter

# Using a conversion context in which the pandas conversion is
# added to the default conversion rules, the rpy2 object
# `mtcars` (an R data frame) is converted to a pandas data frame.
with localconverter(default_converter + pandas2ri.converter) as cv:
    pd_mtcars = mtcars_env['mtcars']
print(type(pd_mtcars))

<class 'pandas.core.frame.DataFrame'>

Using a local converter lets us also go from the pandas data frame to our dplyr-augmented R data frame and use the dplyr transformations on it.

with localconverter(default_converter + pandas2ri.converter) as cv:
    dataf = (DataFrame(pd_mtcars).
             filter('gear>=3').
             mutate(powertoweight='hp*36/wt').
             group_by('gear').
             summarize(mean_ptw='mean(powertoweight)'))

dataf

Reuse. Get things done. Don't reimplement.

		gear	mean_ptw
0	1	4.0	1237.1266499803169
1	2	5.0	2574.0331639315027

		gear	mean_ptw	mean_np_ptw
0	1	4.0	1237.1266499803169	1237.1266499803169
1	2	5.0	2574.0331639315027	2574.0331639315027

		gear	mean_ptw	mean_np_ptw
0	1	4.0	1237.1266499803169	1237.1266499803169
1	2	5.0	2574.0331639315027	2574.0331639315027