Cleanup y_predictions in score of Pipeline functions

[angela97lin]

Cleans up some code in score of pipelines

[codecov]

Codecov Report

Merging #711 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #711   +/-   ##
=======================================
  Coverage   99.07%   99.07%           
=======================================
  Files         140      140           
  Lines        4992     4992           
=======================================
  Hits         4946     4946           
  Misses         46       46           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4c40d8e...4c40d8e. Read the comment docs.

[dsherry]

The change looks good. Will approve once we figure out that codecov thing

[dsherry]

Looks great!

@angela97lin I see the following when I run git grep -ni "notimplemented" on master:

.coveragerc:7:    raise NotImplementedError
docs/source/changelog.rst:227:    * Pipelines now require an estimator as the last component in ``component_list``. Slicing pipelines now throws an ``NotImplementedError`` to avoid returning pipelines without an estimator.
evalml/objectives/objective_base.py:12:        raise NotImplementedError("This objective must have a `name` attribute as a class variable")
evalml/objectives/objective_base.py:19:        raise NotImplementedError("This objective must have a `greater_is_better` boolean attribute as a class variable")
evalml/objectives/objective_base.py:27:        raise NotImplementedError("This objective must have a `score_needs_proba` boolean attribute as a class variable")
evalml/objectives/objective_base.py:41:        raise NotImplementedError("`objective_function` must be implemented.")
evalml/pipelines/components/component_base.py:20:        return NotImplementedError("This component must have `name` as a class variable.")
evalml/pipelines/components/component_base.py:27:        return NotImplementedError("This component must have `model_family` as a class variable.")
evalml/pipelines/components/estimators/estimator.py:14:        return NotImplementedError("This component must have `supported_problem_types` as a class variable.")
evalml/pipelines/pipeline_base.py:36:        return NotImplementedError("This pipeline must have `component_graph` as a class variable.")
evalml/pipelines/pipeline_base.py:118:            raise NotImplementedError('Slicing pipelines is currently not supported.')
evalml/pipelines/pipeline_base.py:125:        raise NotImplementedError('Setting pipeline components is not supported.')
evalml/tests/component_tests/test_components.py:144:            raise NotImplementedError()
evalml/tests/pipeline_tests/test_pipelines.py:184:    with pytest.raises(NotImplementedError, match=setting_err_msg):
evalml/tests/pipeline_tests/test_pipelines.py:188:    with pytest.raises(NotImplementedError, match=slicing_err_msg):
evalml/tuners/tuner.py:20:        raise NotImplementedError
evalml/tuners/tuner.py:33:        raise NotImplementedError
evalml/tuners/tuner.py:42:        raise NotImplementedError

If we're going to go with this pattern, which I'm liking, we should make this change for all relevant callsites, right? It would be great if you did that in this PR, but filing an issue to track a separate PR is fine too.