Tensorflow成本图单位

Question

我正在考虑使用this方法来计算使用sess.run的运行操作的计算复杂性，metadata.cost_graph中compute_cost的单位是什么？

例如，运行

tf.reset_default_graph()

y = tf.matmul([[37.0, -23.0], [1.0, 4.0]], tf.random_uniform([2, 2]))

options = tf.RunOptions()
options.trace_level = tf.RunOptions.FULL_TRACE
ops = tf.GraphOptions(build_cost_model=1)

metadata = tf.RunMetadata()

with tf.Session(config=tf.ConfigProto(log_device_placement=True,
    graph_options=ops)) as sess:

    sess.run(y, options=options, run_metadata=metadata)
    print(metadata.cost_graph)

收益

...
node {
name: "MatMul/a"
device: "/job:localhost/replica:0/task:0/device:CPU:0"
id: 2
output_info {
    size: 16
    alias_input_port: -1
    shape {
    dim {
        size: 2
    }
    dim {
        size: 2
    }
    }
    dtype: DT_FLOAT
}
control_input: 0
compute_cost: 3
persistent_memory_size: 16
}
...