@@ -688,6 +688,55 @@ def _simple_new(
688688 new ._dtype = dtype
689689 return new
690690
691+ @classmethod
692+ def from_spmatrix (cls , data ):
693+ """
694+ Create a SparseArray from a scipy.sparse matrix.
695+
696+ .. versionadded:: 0.25.0
697+
698+ Parameters
699+ ----------
700+ data : scipy.sparse.sp_matrix
701+ This should be a SciPy sparse matrix where the size
702+ of the second dimension is 1. In other words, a
703+ sparse matrix with a single column.
704+
705+ Returns
706+ -------
707+ SparseArray
708+
709+ Examples
710+ --------
711+ >>> import scipy.sparse
712+ >>> mat = scipy.sparse.coo_matrix((4, 1))
713+ >>> pd.SparseArray.from_spmatrix(mat)
714+ [0.0, 0.0, 0.0, 0.0]
715+ Fill: 0.0
716+ IntIndex
717+ Indices: array([], dtype=int32)
718+ """
719+ length , ncol = data .shape
720+
721+ if ncol != 1 :
722+ raise ValueError (
723+ "'data' must have a single column, not '{}'" .format (ncol )
724+ )
725+
726+ # our sparse index classes require that the positions be strictly
727+ # increasing. So we need to sort loc, and arr accordingly.
728+ arr = data .data
729+ idx , _ = data .nonzero ()
730+ loc = np .argsort (idx )
731+ arr = arr .take (loc )
732+ idx .sort ()
733+
734+ zero = np .array (0 , dtype = arr .dtype ).item ()
735+ dtype = SparseDtype (arr .dtype , zero )
736+ index = IntIndex (length , idx )
737+
738+ return cls ._simple_new (arr , index , dtype )
739+
691740 def __array__ (self , dtype = None , copy = True ):
692741 fill_value = self .fill_value
693742
@@ -1899,27 +1948,32 @@ def _make_index(length, indices, kind):
18991948# ----------------------------------------------------------------------------
19001949# Accessor
19011950
1951+
1952+ class BaseAccessor (object ):
1953+ _validation_msg = "Can only use the '.sparse' accessor with Sparse data."
1954+
1955+ def __init__ (self , data = None ):
1956+ self ._parent = data
1957+ self ._validate (data )
1958+
1959+ def _validate (self , data ):
1960+ raise NotImplementedError
1961+
1962+
19021963@delegate_names (SparseArray , ['npoints' , 'density' , 'fill_value' ,
19031964 'sp_values' ],
19041965 typ = 'property' )
1905- class SparseAccessor (PandasDelegate ):
1966+ class SparseAccessor (BaseAccessor , PandasDelegate ):
19061967 """
19071968 Accessor for SparseSparse from other sparse matrix data types.
19081969 """
19091970
1910- def __init__ (self , data = None ):
1911- self ._validate (data )
1912- # Store the Series since we need that for to_coo
1913- self ._parent = data
1914-
1915- @staticmethod
1916- def _validate (data ):
1971+ def _validate (self , data ):
19171972 if not isinstance (data .dtype , SparseDtype ):
1918- msg = "Can only use the '.sparse' accessor with Sparse data."
1919- raise AttributeError (msg )
1973+ raise AttributeError (self ._validation_msg )
19201974
19211975 def _delegate_property_get (self , name , * args , ** kwargs ):
1922- return getattr (self ._parent .values , name )
1976+ return getattr (self ._parent .array , name )
19231977
19241978 def _delegate_method (self , name , * args , ** kwargs ):
19251979 if name == 'from_coo' :
@@ -2033,3 +2087,188 @@ def to_coo(self, row_levels=(0, ), column_levels=(1, ), sort_labels=False):
20332087 column_levels ,
20342088 sort_labels = sort_labels )
20352089 return A , rows , columns
2090+
2091+ def to_dense (self ):
2092+ """
2093+ Convert a Series from sparse values to dense.
2094+
2095+ .. versionadded:: 0.25.0
2096+
2097+ Returns
2098+ -------
2099+ Series:
2100+ A Series with the same values, stored as a dense array.
2101+
2102+ Examples
2103+ --------
2104+ >>> series = pd.Series(pd.SparseArray([0, 1, 0]))
2105+ >>> series
2106+ 0 0
2107+ 1 1
2108+ 2 0
2109+ dtype: Sparse[int64, 0]
2110+
2111+ >>> series.sparse.to_dense()
2112+ 0 0
2113+ 1 1
2114+ 2 0
2115+ dtype: int64
2116+ """
2117+ from pandas import Series
2118+ return Series (self ._parent .array .to_dense (),
2119+ index = self ._parent .index ,
2120+ name = self ._parent .name )
2121+
2122+
2123+ class SparseFrameAccessor (BaseAccessor , PandasDelegate ):
2124+ """
2125+ DataFrame accessor for sparse data.
2126+
2127+ .. versionadded :: 0.25.0
2128+ """
2129+
2130+ def _validate (self , data ):
2131+ dtypes = data .dtypes
2132+ if not all (isinstance (t , SparseDtype ) for t in dtypes ):
2133+ raise AttributeError (self ._validation_msg )
2134+
2135+ @classmethod
2136+ def from_spmatrix (cls , data , index = None , columns = None ):
2137+ """
2138+ Create a new DataFrame from a scipy sparse matrix.
2139+
2140+ .. versionadded:: 0.25.0
2141+
2142+ Parameters
2143+ ----------
2144+ data : scipy.sparse.spmatrix
2145+ Must be convertible to csc format.
2146+ index, columns : Index, optional
2147+ Row and column labels to use for the resulting DataFrame.
2148+ Defaults to a RangeIndex.
2149+
2150+ Returns
2151+ -------
2152+ DataFrame
2153+ Each column of the DataFrame is stored as a
2154+ :class:`SparseArray`.
2155+
2156+ Examples
2157+ --------
2158+ >>> import scipy.sparse
2159+ >>> mat = scipy.sparse.eye(3)
2160+ >>> pd.DataFrame.sparse.from_spmatrix(mat)
2161+ 0 1 2
2162+ 0 1.0 0.0 0.0
2163+ 1 0.0 1.0 0.0
2164+ 2 0.0 0.0 1.0
2165+ """
2166+ from pandas import DataFrame
2167+
2168+ data = data .tocsc ()
2169+ index , columns = cls ._prep_index (data , index , columns )
2170+ sparrays = [
2171+ SparseArray .from_spmatrix (data [:, i ])
2172+ for i in range (data .shape [1 ])
2173+ ]
2174+ data = dict (zip (columns , sparrays ))
2175+ return DataFrame (data , index = index )
2176+
2177+ def to_dense (self ):
2178+ """
2179+ Convert a DataFrame with sparse values to dense.
2180+
2181+ .. versionadded:: 0.25.0
2182+
2183+ Returns
2184+ -------
2185+ DataFrame
2186+ A DataFrame with the same values stored as dense arrays.
2187+
2188+ Examples
2189+ --------
2190+ >>> df = pd.DataFrame({"A": pd.SparseArray([0, 1, 0])})
2191+ >>> df.sparse.to_dense()
2192+ A
2193+ 0 0
2194+ 1 1
2195+ 2 0
2196+ """
2197+ from pandas import DataFrame
2198+
2199+ data = {k : v .array .to_dense ()
2200+ for k , v in compat .iteritems (self ._parent )}
2201+ return DataFrame (data ,
2202+ index = self ._parent .index ,
2203+ columns = self ._parent .columns )
2204+
2205+ def to_coo (self ):
2206+ """
2207+ Return the contents of the frame as a sparse SciPy COO matrix.
2208+
2209+ .. versionadded:: 0.20.0
2210+
2211+ Returns
2212+ -------
2213+ coo_matrix : scipy.sparse.spmatrix
2214+ If the caller is heterogeneous and contains booleans or objects,
2215+ the result will be of dtype=object. See Notes.
2216+
2217+ Notes
2218+ -----
2219+ The dtype will be the lowest-common-denominator type (implicit
2220+ upcasting); that is to say if the dtypes (even of numeric types)
2221+ are mixed, the one that accommodates all will be chosen.
2222+
2223+ e.g. If the dtypes are float16 and float32, dtype will be upcast to
2224+ float32. By numpy.find_common_type convention, mixing int64 and
2225+ and uint64 will result in a float64 dtype.
2226+ """
2227+ try :
2228+ from scipy .sparse import coo_matrix
2229+ except ImportError :
2230+ raise ImportError ('Scipy is not installed' )
2231+
2232+ dtype = find_common_type (self ._parent .dtypes )
2233+ if isinstance (dtype , SparseDtype ):
2234+ dtype = dtype .subtype
2235+
2236+ cols , rows , datas = [], [], []
2237+ for col , name in enumerate (self ._parent ):
2238+ s = self ._parent [name ]
2239+ row = s .array .sp_index .to_int_index ().indices
2240+ cols .append (np .repeat (col , len (row )))
2241+ rows .append (row )
2242+ datas .append (s .array .sp_values .astype (dtype , copy = False ))
2243+
2244+ cols = np .concatenate (cols )
2245+ rows = np .concatenate (rows )
2246+ datas = np .concatenate (datas )
2247+ return coo_matrix ((datas , (rows , cols )), shape = self ._parent .shape )
2248+
2249+ @property
2250+ def density (self ):
2251+ """
2252+ Ratio of non-sparse points to total (dense) data points
2253+ represented in the DataFrame.
2254+ """
2255+ return np .mean ([column .array .density
2256+ for _ , column in self ._parent .iteritems ()])
2257+
2258+ @staticmethod
2259+ def _prep_index (data , index , columns ):
2260+ import pandas .core .indexes .base as ibase
2261+
2262+ N , K = data .shape
2263+ if index is None :
2264+ index = ibase .default_index (N )
2265+ if columns is None :
2266+ columns = ibase .default_index (K )
2267+
2268+ if len (columns ) != K :
2269+ raise ValueError ('Column length mismatch: {columns} vs. {K}'
2270+ .format (columns = len (columns ), K = K ))
2271+ if len (index ) != N :
2272+ raise ValueError ('Index length mismatch: {index} vs. {N}'
2273+ .format (index = len (index ), N = N ))
2274+ return index , columns
0 commit comments