DOC: Complete R interface section

Author: sinhrks

doc/source/r_interface.rst

@@ -6,6 +6,11 @@
    :suppress:
 
    from pandas import *
+   import numpy as np
+   np.random.seed(123456)
+   import matplotlib.pyplot as plt
+   plt.close('all')
+   options.display.mpl_style = 'default'
    options.display.max_rows=15
 
 
@@ -71,24 +76,26 @@ Converting DataFrames into R objects
 .. versionadded:: 0.8
 
 Starting from pandas 0.8, there is **experimental** support to convert
-DataFrames into the equivalent R object (that is, **data.frame**):
+``DataFrame`` into the equivalent R object (that is, **data.frame**) using ``convert_to_r_dataframe`` function:
 
-.. ipython:: python
 
-   from pandas import DataFrame
+.. ipython:: python
 
    df = DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6], 'C':[7,8,9]},
                   index=["one", "two", "three"])
    r_dataframe = com.convert_to_r_dataframe(df)
 
-   print(type(r_dataframe))
+   type(r_dataframe)
    print(r_dataframe)
 
+   print(r_dataframe.rownames)
+   print(r_dataframe.colnames)
+
 
-The DataFrame's index is stored as the ``rownames`` attribute of the
-data.frame instance.
+The ``rpy2.robjects.vectors.DataFrame`` index is stored as the ``rownames``,  and columns are stored as the
+``colnames`` attributes.
 
-You can also use **convert_to_r_matrix** to obtain a ``Matrix`` instance, but
+You can also use ``convert_to_r_matrix`` to obtain a ``rpy2.robjects.vectors.Matrix`` instance, but
 bear in mind that it will only work with homogeneously-typed DataFrames (as
 R matrices bear no information on the data type):
 
@@ -97,14 +104,186 @@ R matrices bear no information on the data type):
 
    r_matrix = com.convert_to_r_matrix(df)
 
-   print(type(r_matrix))
+   type(r_matrix)
    print(r_matrix)
 
 
 Calling R functions with pandas objects
 ---------------------------------------
 
+It is easier to use ``rpy2.robjects`` directly to call R functions.
+You can retrieve R object (including R function) from R namespace by dictionary access of ``robjects.r``.
+
+Below example shows to retrieve R's **sum** function and pass ``rpy2.robjects.vector.DataFrame``.
+Note that the returned value from R **sum** is stored as ``robjects.vectors.Vectors`` type.
+Thus, specify index to get raw values.
+
+See `RPy2 documentation <http://rpy.sourceforge.net/rpy2/doc-2.2/html/index.html>`__ for more.
+
+
+.. ipython:: python
+
+   import rpy2.robjects as robjects
+
+   rsum = robjects.r['sum']
+   rsum_result = rsum(r_dataframe)
+
+   type(rsum_result)
+   rsum_result[0]
+
+
+Preparing Data for R
+--------------------
+
+Load Iris dataset and convert it to R **data.frame**.
+You can pass ``rpy2.robjects.vectors.DataFrame`` to R namespace using ``rpy2.robjects.r.assign``.
+In following examle, `r_iris` DataFrame can be refered as `iris` on R namespace.
+
+
+.. ipython:: python
+
+   iris = com.load_data('iris')
+   iris.head()
+
+   r_iris = com.convert_to_r_dataframe(iris)
+   robjects.r.assign('iris', r_iris);
+
+
+You can convert each data type using R functions if required.
+Function calling ``objects.r`` will execure a passed formula on R's namespace.
+For example, we can check the data type using R's **str** function,
+then convert "Species" column to categorical type (Factor) using R's **factor** function.
+
+
+.. ipython:: python
+
+   print(robjects.r('str(iris)'))
+
+   robjects.r('iris$Species <- factor(iris$Species)');
+   print(robjects.r('str(iris)'))
 
 
 High-level interface to R estimators
 ------------------------------------
+
+Use "setosa" data in iris data set to perform Linear Regression.
+It is much easier to prepare and slice data on pandas side, then convert it to R **data.frame**.
+
+
+.. ipython:: python
+
+   setosa = iris[iris['Species'] == 'setosa']
+   setosa.head()
+
+   r_setosa = com.convert_to_r_dataframe(setosa)
+   robjects.r.assign('setosa', r_setosa);
+
+
+Once DataFrame is passed to R namespace, you can execute R formula to perform Liner Regression.
+
+
+.. ipython:: python
+
+   robjects.r('result <- lm(Sepal.Length~Sepal.Width, data=setosa)');
+   print(robjects.r('summary(result)'))
+
+
+You can retrieve the result from R namespace to python namespace via ``rpy2.robjects.r``.
+If a returned value is R named list, you can check the list of keys via ``names`` attribute.
+To get raw values, access each element specifying index.
+
+
+.. ipython:: python
+
+   result = robjects.r['result']
+
+   print(result.names)
+   print(result.rx('coefficients'))
+
+   intercept, coef1 = result.rx('coefficients')[0]
+   intercept
+   coef1
+
+
+``convert_robj`` function converts retrieved data to python friendly data type.
+In below example, retrieved R **data.frame** of fitted values and confidence interval will be
+converted to pandas ``DataFrame``.
+
+
+.. ipython:: python
+
+   robjects.r('predicted <- predict(result, setosa, interval="prediction")');
+   print(robjects.r('head(predicted)'))
+
+   predicted = robjects.r['predicted']
+   type(predicted)
+
+   predicted = com.convert_robj(predicted)
+   type(predicted)
+   predicted.head()
+
+
+Handling Time Series
+--------------------
+
+Currently, there is no easy way to create R's built-in **ts** object from pandas time series.
+Also, ``Series`` cannot be converted using ``convert_to_r_dataframe`` function.
+Thus, you must create ``rpy2.robjects.vectors.Vector`` instance manually before calling ``robjects.r.assign``.
+
+Use corresponding ``Vector`` class depending on the intended data type.
+See the rpy2 documentation `Vectors and arrays <http://rpy.sourceforge.net/rpy2/doc-2.2/html/vector.html>`__ for more.
+
+Once the ``Vector`` is passed to R's namespace, call R's **ts** function to create **ts** object.
+
+
+.. ipython:: python
+
+   idx = date_range(start='2013-01-01', freq='M', periods=48)
+   vts = Series(np.random.randn(48), index=idx).cumsum()
+   vts
+
+   r_values = robjects.FloatVector(vts.values)
+   robjects.r.assign('values', r_values);
+
+   robjects.r('vts <- ts(values, start=c(2013, 1, 1), frequency=12)');
+   print(robjects.r['vts'])
+
+
+Below example performs Seasonal Decomposition using R's **stl** function, and get the result as `converted` ``DataFrame``.
+Because R's **ts** index cannot be retrieved by ``convert_robj``, assign ``DatetimeIndex`` manually after retrieval.
+
+
+.. ipython:: python
+
+   robjects.r('result <- stl(vts, s.window=12)');
+   result = robjects.r['result']
+
+   print(result.names)
+
+   result_ts = result.rx('time.series')[0]
+   converted = com.convert_robj(result_ts)
+   converted.head()
+
+   converted.index = idx
+   converted.head()
+
+
+Now you have pandas ``DataFrame``, you can perform further operation easily.
+
+
+.. ipython:: python
+
+   fig, axes = plt.subplots(4, 1)
+
+   axes[0].set_ylabel('Original');
+   ax = vts.plot(ax=axes[0]);
+   axes[1].set_ylabel('Trend');
+   ax = converted['trend'].plot(ax=axes[1]);
+
+   axes[2].set_ylabel('Seasonal');
+   ax = converted['seasonal'].plot(ax=axes[2]);
+
+   axes[3].set_ylabel('Residuals');
+   @savefig rpy2_timeseries.png
+   converted['remainder'].plot(ax=axes[3])
+