Two import bug fixes (fixed by Olivier B)

* ValueError: ('invalid number of arguments', mul(Subtensor{0}.0, 18)) This error happened after serializing a model. Scalar ops taking a variable number of arguments had their ufunc set to that op with -1 input parameters, and nout output params. Code has been added to check this condition and set ufunc to None which postpones the building of the ufunc to when we know the actual number of params (or something like that -see OlivierB.). * Optimization bug which seemed rather serious. Fixed bug with Canonizer and DimShuffle Symptom: ValueError: ('Input dimensions do not match.'

Two import bug fixes (fixed by Olivier B)
3c34cf73 · desjagui@atchoum.iro.umontreal.ca · 46ff0418 · 3c34cf73 · 3c34cf73
--- a/theano/tensor/elemwise.py
+++ b/theano/tensor/elemwise.py
@@ -335,7 +335,10 @@ class Elemwise(Op):
    
    def __setstate__(self, d):
        self.__dict__.update(d)
-        self.ufunc = numpy.frompyfunc(self.scalar_op.impl, self.scalar_op.nin, self.scalar_op.nout)
+        if self.scalar_op.nin > 0:
+            self.ufunc = numpy.frompyfunc(self.scalar_op.impl, self.scalar_op.nin, self.scalar_op.nout)
+        else:
+            self.ufunc = None

    def make_node(self, *inputs):
        """
@@ -498,7 +501,13 @@ class Elemwise(Op):
        # the first (faster) version leads to segfaults
        ufunc_args = inputs # + output_storage
        ufunc = self.ufunc or numpy.frompyfunc(self.scalar_op.impl, len(inputs), self.scalar_op.nout)
-        results = ufunc(*ufunc_args)
+        
+        try:
+            results = ufunc(*ufunc_args)
+        except:
+            errormsg = 'Failed calling ufunc for op', self.scalar_op,\
+                        'for params of shape', [arg.shape for arg in ufunc_args]
+            raise Exception, errormsg
        if ufunc.nout == 1: results = [results]
        for result, storage in zip(results, output_storage):
            if storage[0].shape:

--- a/theano/tensor/opt.py
+++ b/theano/tensor/opt.py
@@ -407,7 +407,14 @@ class Canonizer(gof.LocalOptimizer):
    def get_num_denum(self, input):
        if input.owner is None or input.owner.op not in [self.main, self.inverse, self.reciprocal]:
            if input.owner and isinstance(input.owner.op, T.DimShuffle):
-                return self.get_num_denum(input.owner.inputs[0])
+                dsn = input.owner
+                dsop = dsn.op
+                dsi0 = dsn.inputs[0]
+                compatible_order = ('x',) * (input.type.ndim - dsi0.type.ndim) + tuple(range(dsi0.type.ndim))
+                if dsop.new_order == compatible_order:
+                    return self.get_num_denum(input.owner.inputs[0])
+                else:
+                    return [input], []
            else:
                return [input], []
        num = []
@@ -507,6 +514,8 @@ class Canonizer(gof.LocalOptimizer):
        elif op == self.reciprocal:
            reorg = len(iops.intersection([self.inverse, self.reciprocal])) != 0

+        assert len(node.outputs) == 1
+
        orig_num, orig_denum = self.get_num_denum(node.outputs[0])
        num, denum = list(orig_num), list(orig_denum)
        num, denum = self.simplify(num, denum)
@@ -522,6 +531,7 @@ class Canonizer(gof.LocalOptimizer):
            #new = T.fill(out, new)
            elem_op = T.Elemwise(scalar.Identity(scalar.specific_out(getattr(scalar, out.type.dtype))))
            new = T.fill(out, elem_op(new))
+
        if new.broadcastable != out.broadcastable:
            #this case is tricky... we need to provide exactly the same kind of broadcastable
            #pattern, but only if legal...
@@ -541,7 +551,10 @@ class Canonizer(gof.LocalOptimizer):
                new = dimshuffle_op(new)

        # if our if's above worked, this should be true. OTW investigate.
-        assert new.type == out.type
+        if new.type != out.type:
+            print >> sys.stderr, 'CANONIZE FAILED: new out = ', new, out
+            assert new.type == out.type
+
        return [new]

    def __str__(self):