dependency analysis issue with 'component.yaml'-based ops

[amygdala]

I built this pipeline to play around with the new load_component_from_url functionality: https://gist.github.com/amygdala/f6ac2224c598e8355763a45ab502cb49
(The pipeline doesn't quite have correct functionality yet, but that's not important to this issue).
Note line 72, using the op2.after(op1) statement.

When I compile this file, I get the following error. To me it looks like the dependency analysis is trying to use the 'human name' for the component rather than its 'name'.
I'm using the 0.1.12 sdk. (If I comment out line 72, it compiles properly).

Traceback (most recent call last):
  File "cmle_train_deploy.py", line 77, in <module>
    compiler.Compiler().compile(cmle_pipeline, __file__ + '.tar.gz')
  File "/Users/amyu/miniconda3/envs/mlpipeline2/lib/python3.6/site-packages/kfp/compiler/compiler.py", line 572, in compile
    workflow = self._compile(pipeline_func)
  File "/Users/amyu/miniconda3/envs/mlpipeline2/lib/python3.6/site-packages/kfp/compiler/compiler.py", line 562, in _compile
    workflow = self._create_pipeline_workflow(args_list_with_defaults, p)
  File "/Users/amyu/miniconda3/envs/mlpipeline2/lib/python3.6/site-packages/kfp/compiler/compiler.py", line 454, in _create_pipeline_workflow
    templates = self._create_templates(pipeline)
  File "/Users/amyu/miniconda3/envs/mlpipeline2/lib/python3.6/site-packages/kfp/compiler/compiler.py", line 416, in _create_templates
    dependencies = self._get_dependencies(pipeline, new_root_group, op_groups)
  File "/Users/amyu/miniconda3/envs/mlpipeline2/lib/python3.6/site-packages/kfp/compiler/compiler.py", line 190, in _get_dependencies
    upstream_op = pipeline.ops[op_name]
KeyError: 'Train a model with Cloud ML Engine'

[Ark-kun]

Thank you for discovering this DSL compiler bug. I've submitted PR fixing it.

[hongye-sun]

Thanks for the quick fix, Alexey.