Support DataFrame in IncrementalSearchCV

[TomAugspurger]

https://stackoverflow.com/questions/60846247/this-estimator-does-not-support-dask-dataframes

import pandas as pd
import dask.dataframe as dd
from distributed import Client
from sklearn.linear_model import SGDClassifier
from dask_ml.model_selection import IncrementalSearchCV

df = pd.DataFrame({"A": [1, 2, 3, 4, 5], "B": [0, 0, 1, 1, 1]})
ddf = dd.from_pandas(df, 2)
X = df[["A"]]
y = df['B']

model = SGDClassifier()
params = dict(alpha=[0.1, 1])
search = IncrementalSearchCV(model, params)

client = Client()

search.fit(X, y)

If we accept dask dataframes

diff --git a/dask_ml/model_selection/_incremental.py b/dask_ml/model_selection/_incremental.py
index 48f43be9..3219f8c1 100644
--- a/dask_ml/model_selection/_incremental.py
+++ b/dask_ml/model_selection/_incremental.py
@@ -447,8 +447,8 @@ class BaseIncrementalSearchCV(ParallelPostFit):
                 )
             )
 
-        X = self._check_array(X)
-        y = self._check_array(y, ensure_2d=False)
+        X = self._check_array(X, accept_dask_dataframe=True)
+        y = self._check_array(y, ensure_2d=False, accept_dask_dataframe=True)
         scorer = check_scoring(self.estimator, scoring=self.scoring)
         return X, y, scorer
 

We fail at

~/sandbox/dask-ml/dask_ml/model_selection/_incremental.py in _fit(self, X, y, **fit_params)
    556     def _fit(self, X, y, **fit_params):
    557         X, y, scorer = self._validate_parameters(X, y)
--> 558         X_train, X_test, y_train, y_test = self._get_train_test_split(X, y)
    559 
    560         results = yield fit(

~/sandbox/dask-ml/dask_ml/model_selection/_incremental.py in _get_train_test_split(self, X, y, **kwargs)
    484         """
    485         if self.test_size is None:
--> 486             test_size = min(0.2, 1 / X.npartitions)
    487         else:
    488             test_size = self.test_size

AttributeError: 'numpy.ndarray' object has no attribute 'npartitions'

[TomAugspurger]

I believe this was fixed by #701.