import dask.distributed
def f(x, y):
return x, y
client = dask.distributed.Client()
client.map(f, [(1, 2), (2, 3)])
不起作用。
[<Future: status: pending, key: f-137239e2f6eafbe900c0087f550bc0ca>,
<Future: status: pending, key: f-64f918a0c730c63955da91694fcf7acc>]
distributed.worker - WARNING - Compute Failed
Function: f
args: ((1, 2))
kwargs: {}
Exception: TypeError("f() missing 1 required positional argument: 'y'",)
distributed.worker - WARNING - Compute Failed
Function: f
args: ((2, 3))
kwargs: {}
Exception: TypeError("f() missing 1 required positional argument: 'y'",)
答案 0 :(得分:2)
您没有足够的签名权-也许文档不清晰(欢迎提出建议)。 Client.map()
为提交的每个任务采用(可变数量的)参数集,而不是单个可迭代的事物。您应该将其表达为
client.map(f, (1, 2), (2, 3))
或者,如果您想更接近原始图案
client.map(f, *[(1, 2), (2, 3)])
答案 1 :(得分:0)
好吧,documentation对此肯定有点令人困惑。而且我找不到一个清楚地证明了这个问题的例子。因此,我将其分解如下:
def test_fn(a, b, c, d, **kwargs):
return a + b + c + d + kwargs["special"]
futures = client.map(test_fn, *[[1, 2, 3, 4], (1, 2, 3, 4), (1, 2, 3, 4), (1, 2, 3, 4)], special=100)
output = [f.result() for f in futures]
# output = [104, 108, 112, 116]
futures = client.map(test_fn, [1, 2, 3, 4], (1, 2, 3, 4), (1, 2, 3, 4), (1, 2, 3, 4), special=100)
output = [f.result() for f in futures]
# output = [104, 108, 112, 116]
注意事项:
test_fn
的“首次”调用将获得a = b = c = d = 1。)**kwargs
(如special
)将传递给该函数。但是所有函数调用的值都相同。现在我考虑一下,这并不奇怪。我认为它只是遵循Python的concurrent.futures.ProcessPoolExecutor.map()签名。
PS。请注意,即使文档中显示“ Returns:
列表,迭代器或期货队列,具体取决于期货的类型
输入”。实际上,您会收到此错误:Dask no longer supports mapping over Iterators or Queues. Consider using a normal for loop and Client.submit