我正在使用tf探查器,op
视图的典型输出,代码为:
profiler.profile(
tf.get_default_graph(),
run_meta=run_metadata,
cmd='op',
options=opts)
将是
Profile:
node name | # float_ops
MatMul 81.40m float_ops (100.00%, 96.29%)
Mean 795.12k float_ops (3.71%, 0.94%)
Mul 648.40k float_ops (2.77%, 0.77%)
Add 548.00k float_ops (2.00%, 0.65%)
Sub 397.51k float_ops (1.36%, 0.47%)
Square 397.51k float_ops (0.89%, 0.47%)
RealDiv 200.00k float_ops (0.42%, 0.24%)
Sum 100.79k float_ops (0.18%, 0.12%)
Neg 50.00k float_ops (0.06%, 0.06%)
ArgMax 900 float_ops (0.00%, 0.00%)
Equal 102 float_ops (0.00%, 0.00%)
Greater 1 float_ops (0.00%, 0.00%)
======================End of Report==========================
但是我真的无法弄清楚输出中的两个百分比是什么意思,我用Google搜索了一段时间,没有幸运,当我试图读取探查器的源代码时,我发现它包裹得太重了......