Merge (did I miss something in sandbox/downsample?

theano/compile/profilemode.py

@@ -65,25 +65,27 @@ class ProfileMode(Mode):
         """
         local_time = self.local_time[0]
         apply_time = self.apply_time
+        apply_call = self.apply_call
         op_time = self.op_time
+        op_call = self.op_call
 
         print ''
         print 'ProfileMode.print_summary()'
         print '---------------------------'
         print ''
         print 'local_time %fs (Time spent running thunks)'% local_time
-        print 'Apply-wise summary: <% of local_time spent at this position> <total of local_time spent at this position> (<Apply position>, <Apply Op name>)'
-        atimes = [(t/local_time, t, (a[0], str(a[1]))) for a, t in apply_time.items()]
+        print 'Apply-wise summary: <% of local_time spent at this position> <total of local_time spent at this position> <nb_call> <Apply position> <Apply Op name>'
+        atimes = [(t/local_time, t, (a[0], str(a[1])), apply_call[a]) for a, t in apply_time.items()]
         atimes.sort()
         atimes.reverse()
         tot=0
-        for f,t,a in atimes[:n_apply_to_print]:
+        for f,t,a,nb_call in atimes[:n_apply_to_print]:
             tot+=t
-            print '   %4.1f%%  %.3fs  %.3fs  %i  %s' % (f*100, tot, t, a[0], a[1])
+            print '   %4.1f%%  %.3fs  %.3fs  %i  %i %s' % (f*100, tot, t, nb_call, a[0], a[1])
         print '   ... (remaining %i Apply instances account for %.2f%%(%.2fs) of the runtime)'\
                 %(max(0, len(atimes)-n_apply_to_print),
-                  sum(f for f, t, a in atimes[n_apply_to_print:])*100,
-                  sum(t for f, t, a in atimes[n_apply_to_print:]))
+                  sum(f for f, t, a, nb_call in atimes[n_apply_to_print:])*100,
+                  sum(t for f, t, a, nb_call in atimes[n_apply_to_print:]))
 
         flops=False
         flops_msg=''
@@ -94,13 +96,13 @@ class ProfileMode(Mode):
                 print '\nHACK WARNING: we print the flops for some OP, but the logic don\' always work. You need to know the internal of Theano to make it work correctly. Otherwise don\'t use!'
                 break
             
-        print '\nOp-wise summary: < of local_time spent on this kind of Op> <cumulative seconds> <self seconds>%s <Op name>'%(flops_msg)
+        print '\nOp-wise summary: < of local_time spent on this kind of Op> <cumulative seconds> <self seconds>%s <nb_call> <Op name>'%(flops_msg)
 
-        otimes = [(t/local_time, t, a, self.op_cimpl[a]) for a, t in op_time.items()]
+        otimes = [(t/local_time, t, a, self.op_cimpl[a], op_call[a]) for a, t in op_time.items()]
         otimes.sort()
         otimes.reverse()
         tot=0
-        for f,t,a,ci in otimes[:n_ops_to_print]:
+        for f,t,a,ci,nb_call in otimes[:n_ops_to_print]:
             tot+=t
             if ci:
               msg = '*'
@@ -110,39 +112,41 @@ class ProfileMode(Mode):
             if hasattr(a,'flops'):
                 m=a.flops*self.op_call[a]/t/1e6
             if flops:
-                print '   %4.1f%%  %.3fs  %.3fs  %s %7.1f %s' % (f*100, tot, t, msg, m,a)
+                print '   %4.1f%%  %.3fs  %.3fs  %s %7.1f %d %s' % (f*100, tot, t, msg, m, nb_call, a)
             else:
                 print '   %4.1f%%  %.3fs  %.3fs  %s %s' % (f*100, tot, t, msg, a)
         print '   ... (remaining %i Ops account for %6.2f%%(%.2fs) of the runtime)'\
                 %(max(0, len(otimes)-n_ops_to_print),
-                  sum(f for f, t, a, ci in otimes[n_ops_to_print:])*100,
-                  sum(t for f, t, a, ci in otimes[n_ops_to_print:]))
+                  sum(f for f, t, a, ci, nb_call in otimes[n_ops_to_print:])*100,
+                  sum(t for f, t, a, ci, nb_call in otimes[n_ops_to_print:]))
         print '(*) Op is running a c implementation'
 
 
         sop_time={}
+        sop_call={}
         sop_c={} #map each op class to Bool. True iff all applies were done in c.
         for a,t in op_time.items():
             sop_time.setdefault(type(a),0)
             sop_time[type(a)]+=t
             sop_c.setdefault(type(a),True)
             sop_c[type(a)]=sop_c[type(a)] and self.op_cimpl[a]
-        print '\nSingle Op-wise summary: <% of local_time spent on this kind of Op> <cumulative seconds> <self seconds> <Op name>'
-        sotimes = [(t/local_time, t, a, sop_c[a]) for a, t in sop_time.items()]
+            sop_call[type(a)]=sop_call.get(type(a),0)+op_call[a]
+        print '\nSingle Op-wise summary: <% of local_time spent on this kind of Op> <cumulative seconds> <self seconds> <nb_call> <Op name>'
+        sotimes = [(t/local_time, t, a, sop_c[a], sop_call[a]) for a, t in sop_time.items()]
         sotimes.sort()
         sotimes.reverse()
         tot=0
-        for f,t,a,ci in sotimes[:n_ops_to_print]:
+        for f,t,a,ci, nb_call in sotimes[:n_ops_to_print]:
             tot+=t
             if ci:
               msg = '*'
             else:
               msg = ' '
-            print '   %4.1f%%  %.3fs  %.3fs  %s %s' % (f*100, tot, t, msg, a)
+            print '   %4.1f%%  %.3fs  %.3fs  %s %d %s' % (f*100, tot, t, msg, nb_call, a)
         print '   ... (remaining %i Ops account for %.2f%%(%.2fs) of the runtime)'\
                 %(max(0, len(sotimes)-n_ops_to_print),
-                  sum(f for f, t, a in sotimes[n_ops_to_print:])*100,
-                  sum(t for f, t, a in sotimes[n_ops_to_print:]))
+                  sum(f for f, t, a, nb_call in sotimes[n_ops_to_print:])*100,
+                  sum(t for f, t, a, nb_call in sotimes[n_ops_to_print:]))
         print '(*) Op is running a c implementation'
         print 'compile time: %.3fs'%self.compile_time
 

theano/gof/link.py

@@ -144,9 +144,9 @@ class Container(object):
         self.storage = storage
         self.readonly = readonly
         self.strict = strict
-    def __get(self):
+    def __get__(self):
         return self.storage[0]
-    def __set(self, value):
+    def __set__(self, value):
         if self.readonly:
             raise Exception("Cannot set readonly storage: %s" % self.name)
         try:
@@ -160,8 +160,8 @@ class Container(object):
         except Exception, e:
             e.args = e.args + (('Container name "%s"' % self.name),)
             raise
-    data = property(__get, __set)
-    value = property(__get, __set)
+    data = property(__get__, __set__)
+    value = property(__get__, __set__)
     def __str__(self):
         return "<" + str(self.storage[0]) + ">"
     def __repr__(self):

theano/tensor/basic.py

@@ -405,9 +405,11 @@ class TensorType(Type):
 
     def c_extract(self, name, sub):
         """Override `CLinkerOp.c_extract` """
+        # TODO: make the error message print out the dtype of the
+        # input received.
         return """
         %(name)s = NULL;
-        type_num_%(name)s = %(type_num)s;
+        type_num_%(name)s = ((PyArrayObject*)py_%(name)s)->descr->type_num; //we expect %(type_num)s
         if (py_%(name)s == Py_None) {
             // We can either fail here or set %(name)s to NULL and rely on Ops using
             // tensors to handle the NULL case, but if they fail to do so they'll end up
@@ -419,7 +421,7 @@ class TensorType(Type):
             PyErr_SetString(PyExc_ValueError, "expected an ndarray");
             %(fail)s
         }
-        else if (((PyArrayObject*)py_%(name)s)->descr->type_num != %(type_num)s) {
+        else if (type_num_%(name)s != %(type_num)s) {
             PyErr_SetString(PyExc_ValueError, "expected %(type_num)s");
             %(fail)s
         }
@@ -1394,6 +1396,15 @@ class Repeat(gof.Op):
 
 repeat = Repeat()
 
+class SetDefault(gof.Op):
+    view_map = {0: [1]}
+    def make_node(self, x, default):
+        assert x.type == default.type
+        return gof.Apply(self, [x, default], [default.type()])
+    def perform(self, node, (x, default), (out, )):
+        out[0] = default.copy() if x is None else x
+
+setdefault = SetDefault()
 
 
 ##########################
@@ -1879,7 +1890,6 @@ class Split(Op):
         return [join(axis, *g_outputs), None, None]
 
 
-
 class Rebroadcast(Op):
     """
     Change the input's broadcastable fields in
@@ -1919,6 +1929,7 @@ def unbroadcast(x, *axes):
     return Rebroadcast(*[(axis, False) for axis in axes])(x)
 
 
+
 class Join(Op):
     """
     Concatenate multiple `TensorVariable`s along some axis.
@@ -1970,6 +1981,7 @@ class Join(Op):
         # for the axis dimension.
         # All concatenated elements must also have the same broadcastable
         # dimensions.
+        orig = as_tensor_variable_args
         if isinstance(axis, int):
             bcasts = [x.type.broadcastable[0:axis] + \
                       x.type.broadcastable[axis + 1:] for x in as_tensor_variable_args]
@@ -1991,7 +2003,9 @@ class Join(Op):
 
         outputs = [tensor(dtype = out_dtype,
                           broadcastable = bcastable)]
-        return Apply(self, inputs, outputs)
+        node = Apply(self, inputs, outputs)
+        node.tag.shape_zero = None if any(not x.type.broadcastable[0] for x in orig) else len(orig)
+        return node
 
     def perform(self, node, axis_and_tensors, (out, )):
         axis, tensors = axis_and_tensors[0], axis_and_tensors[1:]
@@ -2031,13 +2045,10 @@ class Join(Op):
         assert isinstance(node.owner.op, Join)
         if node.ndim != 1:
             raise TypeError('argument must be symbolic vector')
-        inputs = node.owner.inputs
-        axis, tensors = inputs[0], inputs[1:]
-        # if v is a vector, then axis must be 0
-        # the question is whether all the inputs are broadcastable.
-        if all(i.broadcastable[0] for i in tensors):
-            return len(tensors)
-        raise ValueError("could not determine vector length")
+        if node.owner.tag.shape_zero is None:
+          raise ValueError("could not determine vector length")
+        else:
+          return node.owner.tag.shape_zero
 
 @_redefine_asRoutine(Join())
 def join(axis, *tensors):
@@ -2137,7 +2148,7 @@ def get_vector_length(v):
     if v.owner and isinstance(v.owner.op, Join):
         try:
             return join.vec_length(v)
-        except:
+        except ValueError:
             pass
     if v.owner and v.owner.op == shape:
         return v.owner.inputs[0].type.ndim
@@ -2554,6 +2565,7 @@ class Outer(Op):
         return "outer"
 outer = Outer()
 
+
 #########################
 # Gradient
 #########################

theano/tensor/nnet.py

@@ -401,10 +401,12 @@ def local_softmax_with_bias(node):
             vectors = []
             non_vectors = []
             for x_in in x.owner.inputs:
-                if list(x_in.type.broadcastable) == [True, False] \
-                        and isinstance(x_in.owner.op, tensor.DimShuffle):
-                    assert len(x_in.owner.inputs)==1
-                    vectors.append(x_in.owner.inputs[0])
+                if list(x_in.type.broadcastable) == [True, False]:
+                    if x_in.owner and isinstance(x_in.owner.op, tensor.DimShuffle):
+                        assert len(x_in.owner.inputs)==1
+                        vectors.append(x_in.owner.inputs[0])
+                    else:
+                        vectors.append(tensor.DimShuffle((True, False), (1,))(x_in))
                 else:
                     non_vectors.append(x_in)