Merge branch 'master' of github.com:pks/cdec-dtrain

1 files changed, 56 insertions, 25 deletions

diff --git a/python/src/vectors.pxi b/python/src/vectors.pxi
index 74d3a0bd..ce95968c 100644
--- a/python/src/vectors.pxi
+++ b/python/src/vectors.pxi
@@ -1,16 +1,20 @@
 from cython.operator cimport preincrement as pinc
 
 cdef class DenseVector:
-    cdef vector[weight_t]* vector
+    cdef vector[weight_t]* vector # Not owned by DenseVector
+
+    def __len__(self):
+        return self.vector.size()
 
     def __getitem__(self, char* fname):
-        cdef unsigned fid = FDConvert(fname)
-        if fid <= self.vector.size():
+        cdef int fid = FDConvert(fname)
+        if 0 <= fid < self.vector.size():
             return self.vector[0][fid]
         raise KeyError(fname)
     
     def __setitem__(self, char* fname, float value):
-        cdef unsigned fid = FDConvert(<char *>fname)
+        cdef int fid = FDConvert(<char *>fname)
+        if fid < 0: raise KeyError(fname)
         if self.vector.size() <= fid:
             self.vector.resize(fid + 1)
         self.vector[0][fid] = value
@@ -32,40 +36,43 @@ cdef class DenseVector:
 cdef class SparseVector:
     cdef FastSparseVector[weight_t]* vector
 
+    def __dealloc__(self):
+        del self.vector
+
+    def copy(self):
+        return self * 1
+
     def __getitem__(self, char* fname):
-        cdef unsigned fid = FDConvert(fname)
+        cdef int fid = FDConvert(fname)
+        if fid < 0: raise KeyError(fname)
         return self.vector.value(fid)
     
     def __setitem__(self, char* fname, float value):
-        cdef unsigned fid = FDConvert(<char *>fname)
+        cdef int fid = FDConvert(<char *>fname)
+        if fid < 0: raise KeyError(fname)
         self.vector.set_value(fid, value)
 
     def __iter__(self):
         cdef FastSparseVector[weight_t].const_iterator* it = new FastSparseVector[weight_t].const_iterator(self.vector[0], False)
-        cdef str fname
-        for i in range(self.vector.size()):
-            fname = FDConvert(it[0].ptr().first).c_str()
-            yield (fname, it[0].ptr().second)
-            pinc(it[0])
+        try:
+            for i in range(self.vector.size()):
+                yield (FDConvert(it[0].ptr().first).c_str(), it[0].ptr().second)
+                pinc(it[0]) # ++it
+        finally:
+            del it
 
     def dot(self, other):
         if isinstance(other, DenseVector):
             return self.vector.dot((<DenseVector> other).vector[0])
         elif isinstance(other, SparseVector):
             return self.vector.dot((<SparseVector> other).vector[0])
-        raise ValueError('cannot take the dot product of %s and SparseVector' % type(other))
-
-    def todense(self):
-        cdef DenseVector dense = DenseVector()
-        dense.vector = new vector[weight_t]()
-        self.vector.init_vector(dense.vector)
-        return dense
+        raise TypeError('cannot take the dot product of %s and SparseVector' % type(other))
     
-    def __richcmp__(SparseVector self, SparseVector other, int op):
+    def __richcmp__(SparseVector x, SparseVector y, int op):
         if op == 2: # ==
-            return self.vector[0] == other.vector[0]
+            return x.vector[0] == y.vector[0]
         elif op == 3: # !=
-            return not (self == other)
+            return not (x == y)
         raise NotImplemented('comparison not implemented for SparseVector')
 
     def __len__(self):
@@ -73,6 +80,12 @@ cdef class SparseVector:
 
     def __contains__(self, char* fname):
         return self.vector.nonzero(FDConvert(fname))
+
+    def __neg__(self):
+        cdef SparseVector result = SparseVector()
+        result.vector = new FastSparseVector[weight_t](self.vector[0])
+        result.vector[0] *= -1.0
+        return result
     
     def __iadd__(SparseVector self, SparseVector other):
         self.vector[0] += other.vector[0]
@@ -90,12 +103,30 @@ cdef class SparseVector:
         self.vector[0] /= scalar
         return self
 
-    def __add__(SparseVector self, SparseVector other):
+    def __add__(SparseVector x, SparseVector y):
+        cdef SparseVector result = SparseVector()
+        result.vector = new FastSparseVector[weight_t](x.vector[0] + y.vector[0])
+        return result
+
+    def __sub__(SparseVector x, SparseVector y):
+        cdef SparseVector result = SparseVector()
+        result.vector = new FastSparseVector[weight_t](x.vector[0] - y.vector[0])
+        return result
+
+    def __mul__(x, y):
+        cdef SparseVector vector
+        cdef float scalar
+        if isinstance(x, SparseVector): vector, scalar = x, y
+        else: vector, scalar = y, x
         cdef SparseVector result = SparseVector()
-        result.vector = new FastSparseVector[weight_t](self.vector[0] + other.vector[0])
+        result.vector = new FastSparseVector[weight_t](vector.vector[0] * scalar)
         return result
 
-    def __sub__(SparseVector self, SparseVector other):
+    def __div__(x, y):
+        cdef SparseVector vector
+        cdef float scalar
+        if isinstance(x, SparseVector): vector, scalar = x, y
+        else: vector, scalar = y, x
         cdef SparseVector result = SparseVector()
-        result.vector = new FastSparseVector[weight_t](self.vector[0] - other.vector[0])
+        result.vector = new FastSparseVector[weight_t](vector.vector[0] / scalar)
         return result