pep8

theano/sandbox/cuda/tests/test_opt.py

@@ -156,7 +156,6 @@ def test_gpualloc():
 
 def test_gpuallocempty():
 
-
     f_gpu = theano.function([], tensor.AllocEmpty('float32')(2,3),
                         mode=mode_with_gpu)
     l_gpu = f_gpu.maker.fgraph.toposort()

theano/tensor/basic.py

@@ -5505,7 +5505,6 @@ class AllocEmpty(gof.Op):
         out, = out_
         sh = tuple([int(i) for i in inputs])
         if out[0] is None or out[0].shape != sh:
-            # XXX: We could implement and call CudaNdarray.empty(sh) instead.
             out[0] = numpy.empty(sh, dtype=self.dtype)
 
     def c_code(self, node, name, inputs, out_, sub):
@@ -5515,10 +5514,10 @@ class AllocEmpty(gof.Op):
         shps = inputs
         nd = len(shps)
         str = "npy_intp dims[%(nd)s];\n" % locals()
-        
         for idx, sh in enumerate(shps):
-            str +="dims[%(idx)s] =" \
-                  "((npy_intp)((dtype_%(sh)s*)PyArray_DATA(%(sh)s))[0]);\n" % locals()
+            str += "dims[%(idx)s] =" \
+                   "((npy_intp)((dtype_%(sh)s*)" \
+                   " PyArray_DATA(%(sh)s))[0]);\n" % locals()
 
         # Validate that the output storage exists
         str += "if(%(out)s==NULL\n" % locals()

theano/tensor/blas.py

@@ -2090,7 +2090,7 @@ def local_dot22_to_dot22scalar(node):
                                                   d.owner.inputs[1].shape[1])
         zero = T.as_tensor_variable(numpy.asarray(0, dtype=a.dtype))
         return [T.mul(gemm(z, a, d.owner.inputs[0], d.owner.inputs[1],
-                                                    zero), *o)]
+                      zero), *o)]
 # must happen after gemm as the gemm optimizer don't understant
 # dot22scalar and gemm give more speed up then dot22scalar
 blas_optdb.register('local_dot22_to_dot22scalar',