Skip to content

P
pytensor
提交 4d85200e authored 作者: gdesjardins's avatar gdesjardins
浏览文件
操作

BUG FIX: when using scan with k=1, tensor outputs (of the Scan op) should not be

left-padded if the user asked NOT to store intermediate results (this should be the default-case for shared variables)
上级 8d2f4abe
隐藏空白字符变更
内嵌 并排
正在显示 包含 1 行增加 和 1 行删除
theano/scan.py
浏览文件 @ 4d85200e
...@@ -862,7 +862,7 @@ def scan(fn, sequences=[], outputs_info=[], non_sequences=[], ...@@ -862,7 +862,7 @@ def scan(fn, sequences=[], outputs_info=[], non_sequences=[],
else: else:
# If we do not actually need scan # If we do not actually need scan
for pos, inner_out in enumerate(inner_fn_outs): for pos, inner_out in enumerate(inner_fn_outs):
if isinstance(inner_out.type, tensor.TensorType): if isinstance(inner_out.type, tensor.TensorType) and store_steps[pos] != 1:
inner_fn_outs[pos] = tensor.unbroadcast( tensor.shape_padleft(inner_out),0) inner_fn_outs[pos] = tensor.unbroadcast( tensor.shape_padleft(inner_out),0)
values = inner_fn_outs values = inner_fn_outs
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论