@@ -689,6 +689,55 @@ def _simple_new(
689689 new ._dtype = dtype
690690 return new
691691
692+ @classmethod
693+ def from_spmatrix (cls , data ):
694+ """
695+ Create a SparseArray from a scipy.sparse matrix.
696+
697+ .. versionadded:: 0.25.0
698+
699+ Parameters
700+ ----------
701+ data : scipy.sparse.sp_matrix
702+ This should be a SciPy sparse matrix where the size
703+ of the second dimension is 1. In other words, a
704+ sparse matrix with a single column.
705+
706+ Returns
707+ -------
708+ SparseArray
709+
710+ Examples
711+ --------
712+ >>> import scipy.sparse
713+ >>> mat = scipy.sparse.coo_matrix((4, 1))
714+ >>> pd.SparseArray.from_spmatrix(mat)
715+ [0.0, 0.0, 0.0, 0.0]
716+ Fill: 0.0
717+ IntIndex
718+ Indices: array([], dtype=int32)
719+ """
720+ length , ncol = data .shape
721+
722+ if ncol != 1 :
723+ raise ValueError (
724+ "'data' must have a single column, not '{}'" .format (ncol )
725+ )
726+
727+ # our sparse index classes require that the positions be strictly
728+ # increasing. So we need to sort loc, and arr accordingly.
729+ arr = data .data
730+ idx , _ = data .nonzero ()
731+ loc = np .argsort (idx )
732+ arr = arr .take (loc )
733+ idx .sort ()
734+
735+ zero = np .array (0 , dtype = arr .dtype ).item ()
736+ dtype = SparseDtype (arr .dtype , zero )
737+ index = IntIndex (length , idx )
738+
739+ return cls ._simple_new (arr , index , dtype )
740+
692741 def __array__ (self , dtype = None , copy = True ):
693742 fill_value = self .fill_value
694743
@@ -1898,27 +1947,32 @@ def _make_index(length, indices, kind):
18981947# ----------------------------------------------------------------------------
18991948# Accessor
19001949
1950+
1951+ class BaseAccessor (object ):
1952+ _validation_msg = "Can only use the '.sparse' accessor with Sparse data."
1953+
1954+ def __init__ (self , data = None ):
1955+ self ._parent = data
1956+ self ._validate (data )
1957+
1958+ def _validate (self , data ):
1959+ raise NotImplementedError
1960+
1961+
19011962@delegate_names (SparseArray , ['npoints' , 'density' , 'fill_value' ,
19021963 'sp_values' ],
19031964 typ = 'property' )
1904- class SparseAccessor (PandasDelegate ):
1965+ class SparseAccessor (BaseAccessor , PandasDelegate ):
19051966 """
19061967 Accessor for SparseSparse from other sparse matrix data types.
19071968 """
19081969
1909- def __init__ (self , data = None ):
1910- self ._validate (data )
1911- # Store the Series since we need that for to_coo
1912- self ._parent = data
1913-
1914- @staticmethod
1915- def _validate (data ):
1970+ def _validate (self , data ):
19161971 if not isinstance (data .dtype , SparseDtype ):
1917- msg = "Can only use the '.sparse' accessor with Sparse data."
1918- raise AttributeError (msg )
1972+ raise AttributeError (self ._validation_msg )
19191973
19201974 def _delegate_property_get (self , name , * args , ** kwargs ):
1921- return getattr (self ._parent .values , name )
1975+ return getattr (self ._parent .array , name )
19221976
19231977 def _delegate_method (self , name , * args , ** kwargs ):
19241978 if name == 'from_coo' :
@@ -2032,3 +2086,188 @@ def to_coo(self, row_levels=(0, ), column_levels=(1, ), sort_labels=False):
20322086 column_levels ,
20332087 sort_labels = sort_labels )
20342088 return A , rows , columns
2089+
2090+ def to_dense (self ):
2091+ """
2092+ Convert a Series from sparse values to dense.
2093+
2094+ .. versionadded:: 0.25.0
2095+
2096+ Returns
2097+ -------
2098+ Series:
2099+ A Series with the same values, stored as a dense array.
2100+
2101+ Examples
2102+ --------
2103+ >>> series = pd.Series(pd.SparseArray([0, 1, 0]))
2104+ >>> series
2105+ 0 0
2106+ 1 1
2107+ 2 0
2108+ dtype: Sparse[int64, 0]
2109+
2110+ >>> series.sparse.to_dense()
2111+ 0 0
2112+ 1 1
2113+ 2 0
2114+ dtype: int64
2115+ """
2116+ from pandas import Series
2117+ return Series (self ._parent .array .to_dense (),
2118+ index = self ._parent .index ,
2119+ name = self ._parent .name )
2120+
2121+
2122+ class SparseFrameAccessor (BaseAccessor , PandasDelegate ):
2123+ """
2124+ DataFrame accessor for sparse data.
2125+
2126+ .. versionadded :: 0.25.0
2127+ """
2128+
2129+ def _validate (self , data ):
2130+ dtypes = data .dtypes
2131+ if not all (isinstance (t , SparseDtype ) for t in dtypes ):
2132+ raise AttributeError (self ._validation_msg )
2133+
2134+ @classmethod
2135+ def from_spmatrix (cls , data , index = None , columns = None ):
2136+ """
2137+ Create a new DataFrame from a scipy sparse matrix.
2138+
2139+ .. versionadded:: 0.25.0
2140+
2141+ Parameters
2142+ ----------
2143+ data : scipy.sparse.spmatrix
2144+ Must be convertible to csc format.
2145+ index, columns : Index, optional
2146+ Row and column labels to use for the resulting DataFrame.
2147+ Defaults to a RangeIndex.
2148+
2149+ Returns
2150+ -------
2151+ DataFrame
2152+ Each column of the DataFrame is stored as a
2153+ :class:`SparseArray`.
2154+
2155+ Examples
2156+ --------
2157+ >>> import scipy.sparse
2158+ >>> mat = scipy.sparse.eye(3)
2159+ >>> pd.DataFrame.sparse.from_spmatrix(mat)
2160+ 0 1 2
2161+ 0 1.0 0.0 0.0
2162+ 1 0.0 1.0 0.0
2163+ 2 0.0 0.0 1.0
2164+ """
2165+ from pandas import DataFrame
2166+
2167+ data = data .tocsc ()
2168+ index , columns = cls ._prep_index (data , index , columns )
2169+ sparrays = [
2170+ SparseArray .from_spmatrix (data [:, i ])
2171+ for i in range (data .shape [1 ])
2172+ ]
2173+ data = dict (zip (columns , sparrays ))
2174+ return DataFrame (data , index = index )
2175+
2176+ def to_dense (self ):
2177+ """
2178+ Convert a DataFrame with sparse values to dense.
2179+
2180+ .. versionadded:: 0.25.0
2181+
2182+ Returns
2183+ -------
2184+ DataFrame
2185+ A DataFrame with the same values stored as dense arrays.
2186+
2187+ Examples
2188+ --------
2189+ >>> df = pd.DataFrame({"A": pd.SparseArray([0, 1, 0])})
2190+ >>> df.sparse.to_dense()
2191+ A
2192+ 0 0
2193+ 1 1
2194+ 2 0
2195+ """
2196+ from pandas import DataFrame
2197+
2198+ data = {k : v .array .to_dense ()
2199+ for k , v in compat .iteritems (self ._parent )}
2200+ return DataFrame (data ,
2201+ index = self ._parent .index ,
2202+ columns = self ._parent .columns )
2203+
2204+ def to_coo (self ):
2205+ """
2206+ Return the contents of the frame as a sparse SciPy COO matrix.
2207+
2208+ .. versionadded:: 0.20.0
2209+
2210+ Returns
2211+ -------
2212+ coo_matrix : scipy.sparse.spmatrix
2213+ If the caller is heterogeneous and contains booleans or objects,
2214+ the result will be of dtype=object. See Notes.
2215+
2216+ Notes
2217+ -----
2218+ The dtype will be the lowest-common-denominator type (implicit
2219+ upcasting); that is to say if the dtypes (even of numeric types)
2220+ are mixed, the one that accommodates all will be chosen.
2221+
2222+ e.g. If the dtypes are float16 and float32, dtype will be upcast to
2223+ float32. By numpy.find_common_type convention, mixing int64 and
2224+ and uint64 will result in a float64 dtype.
2225+ """
2226+ try :
2227+ from scipy .sparse import coo_matrix
2228+ except ImportError :
2229+ raise ImportError ('Scipy is not installed' )
2230+
2231+ dtype = find_common_type (self ._parent .dtypes )
2232+ if isinstance (dtype , SparseDtype ):
2233+ dtype = dtype .subtype
2234+
2235+ cols , rows , datas = [], [], []
2236+ for col , name in enumerate (self ._parent ):
2237+ s = self ._parent [name ]
2238+ row = s .array .sp_index .to_int_index ().indices
2239+ cols .append (np .repeat (col , len (row )))
2240+ rows .append (row )
2241+ datas .append (s .array .sp_values .astype (dtype , copy = False ))
2242+
2243+ cols = np .concatenate (cols )
2244+ rows = np .concatenate (rows )
2245+ datas = np .concatenate (datas )
2246+ return coo_matrix ((datas , (rows , cols )), shape = self ._parent .shape )
2247+
2248+ @property
2249+ def density (self ):
2250+ """
2251+ Ratio of non-sparse points to total (dense) data points
2252+ represented in the DataFrame.
2253+ """
2254+ return np .mean ([column .array .density
2255+ for _ , column in self ._parent .iteritems ()])
2256+
2257+ @staticmethod
2258+ def _prep_index (data , index , columns ):
2259+ import pandas .core .indexes .base as ibase
2260+
2261+ N , K = data .shape
2262+ if index is None :
2263+ index = ibase .default_index (N )
2264+ if columns is None :
2265+ columns = ibase .default_index (K )
2266+
2267+ if len (columns ) != K :
2268+ raise ValueError ('Column length mismatch: {columns} vs. {K}'
2269+ .format (columns = len (columns ), K = K ))
2270+ if len (index ) != N :
2271+ raise ValueError ('Index length mismatch: {index} vs. {N}'
2272+ .format (index = len (index ), N = N ))
2273+ return index , columns
0 commit comments