[python] Explicit candidate set; memory leak fixes

1 files changed, 24 insertions, 17 deletions

diff --git a/python/src/vectors.pxi b/python/src/vectors.pxi
index 233c9530..fc0c365f 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
@@ -32,21 +36,30 @@ 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):
@@ -54,12 +67,6 @@ cdef class SparseVector:
         elif isinstance(other, SparseVector):
             return self.vector.dot((<SparseVector> other).vector[0])
         raise TypeError('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
     
     def __richcmp__(SparseVector x, SparseVector y, int op):
         if op == 2: # ==