提交 3ae55342 authored 作者: Razvan Pascanu's avatar Razvan Pascanu

Optimization removed due to a bug noticed in the tutorials .. I will be trying…

Optimization removed due to a bug noticed in the tutorials .. I will be trying to fix this as soon as possible.
上级 2e7311dc
......@@ -1498,6 +1498,7 @@ class ScanRemoveFromGraph(Optimizer):
for idx in xrange(len(my_outs)):
t = my_outs[idx]
nwout = tensor.Rebroadcast((0,False))(tensor.shape_padleft(t))
print 'replacing', node.outputs[idx], nwout
env.replace(node.outputs[idx],nwout)
# we are done ...
......@@ -1505,4 +1506,4 @@ class ScanRemoveFromGraph(Optimizer):
# is 30 soon enough !? I want to do it as early as possible .. such that
# the new graph gets optimized
optdb.register('scanOp_remove_from_graph', ScanRemoveFromGraph() , 30, 'fast_run')
#optdb.register('scanOp_remove_from_graph', ScanRemoveFromGraph() , 30, 'fast_run')
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论