Description

TypeError: 'tuple' object does not support item assignment error thrown when calling fit on a Pipeline whose steps parameter was initialized with an n-tuple (as opposed to a list) of (name, transform) tuples.

Previous versions of sklearn Pipelines worked when an n-tuple was used for steps. I guess new functionality associated with Pipelines now requires that steps be mutable.

I guess there are a few choices here:

1. Automatically handle the case of an n-tuple being passed by creating a new list for self.steps and copying over the tuple's items to it. And maybe throw a warning for behavior's deprecation.
2. Throw an immediate error during initialization if steps is a tuple. Of course, this is still a breaking change, but at least, the user won't have to wait until calling fit before discovering something is wrong.

I could submit a PR for this if you want.

Steps/Code to Reproduce

from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from sklearn.decomposition import PCA
from sklearn.datasets import load_breast_cancer

pipeline = Pipeline((('pca', PCA()), ('rf', RandomForestClassifier())))
X, y = load_breast_cancer(True)
pipeline.fit(X, y)

Expected Results

This had no errors in previous versions

Actual Results

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/dist-packages/sklearn/pipeline.py", line 257, in fit
    Xt, fit_params = self._fit(X, y, **fit_params)
  File "/usr/local/lib/python2.7/dist-packages/sklearn/pipeline.py", line 226, in _fit
    self.steps[step_idx] = (name, fitted_transformer)
TypeError: 'tuple' object does not support item assignment

Versions

Linux-4.4.0-36-generic-x86_64-with-Ubuntu-14.04-trusty
('Python', '2.7.6 (default, Jun 22 2015, 17:58:13) \n[GCC 4.8.2]')
('NumPy', '1.13.1')
('SciPy', '0.19.0')
('Scikit-Learn', '0.19.0')