如何在元流中创建嵌套分支？

Question

我用来metaflow创建一个文本处理管道，如下所示：-

                                 ___F------
                     ______ D---|          |  
                    |           |___G---|  |__>  
          ____B-----|                   |----->H
         |          |______E_________________> ^
      A -|                                     |
         |____C________________________________|

根据文档，branch允许并行计算步骤，并用于并行计算 (B, C)、(D, E) 和 (F, G)。最后所有分支都在 H 处连接。以下是实现此逻辑的代码：-

from metaflow import FlowSpec, step

class TextProcessing(FlowSpec):

  @step
  def a(self):
    ....

    self.next(self.b, self.c)

  @step
  def c(self):
    result1 = {}

    ....

    self.next(self.join)

  @step
  def b(self):
    ....

    self.next(self.d, self.e)

  @step
  def e(self):
    result2 = []
    .....

    self.next(self.join)

  @step
  def d(self):
    ....

    self.next(self.f, self.g)

  @step
  def f(self):
    result3 = []
    ....

    self.next(self.join)

  @step
  def g(self):
    result4 = []
    .....

    self.next(self.join)


  @step
  def join(self, results):
    data = [results.c.result, results.e.result2, result.f.result3, result.g.result4]
    print(data)

    self.next(self.end)

  @step
  def end(self):
    pass

etl = TextProcessing()

在运行时python main.py run，我收到以下错误：-

Metaflow 2.2.10 executing TextProcessing for user:ubuntu
Validating your flow...
    Validity checker found an issue on line 83:
    Step join seems like a join step (it takes an extra input argument) but an incorrect number of steps (c, e, f, g) lead to it. This join was expecting 2 incoming paths, starting from splitted step(s) f, g.

有人可以指出我哪里出错了吗？

Answer 1

再次仔细阅读文档后，我意识到我没有正确处理连接。根据文档metaflow-2.2.10：-

请注意，您可以任意嵌套分支，即可以在分支内部进行分支。只需记住加入您创建的所有分支即可。

这意味着每个分支都应该加入。为了连接来自分支的值，metaflow提供merge_artifacts实用函数来帮助传播明确的值。

由于工作流程中有三个分支，因此添加了三个联接步骤来合并结果。

以下更改对我有用：-

from metaflow import FlowSpec, step

class TextProcessing(FlowSpec):

  @step
  def a(self):
    ....

    self.next(self.b, self.c)

  @step
  def c(self):
    result1 = {}

    ....

    self.next(self.merge_3)

  @step
  def b(self):
    ....

    self.next(self.d, self.e)

  @step
  def e(self):
    result2 = []
    .....

    self.next(self.merge_2)

  @step
  def d(self):
    ....

    self.next(self.f, self.g)

  @step
  def f(self):
    result3 = []
    ....

    self.next(self.merge_1)

  @step
  def g(self):
    result4 = []
    .....

    self.next(self.merge_1)

  @step
  def merge_1(self, results):
    self.result = {
      'result4' : results.g.result4,
      'result3' : results.f.result3
    }

    self.next(self.merge_2)

  @step
  def merge_2(self, results):
    self.result = { 'result2' : results.e.result2, **results.merge_1.result }
    self.merge_artifacts(results, include=['result'])
    self.next(self.merge_3)

  @step
  def merge_3(self, results):
    self.result = { 'c' : results.c.result1, **results.merge_2.result }
    self.merge_artifacts(results, include=['result'])
    self.next(self.end)

  @step
  def end(self):
    print(self.result)

etl = TextProcessing()