input validation in predict of DummyRegressor is too strict

[dcatteeu]

Description

sklearn.dummy.DummyRegressor requires X to be numeric and have at least 1 feature in predict. IMO, this is too strict.

Also, it is not done in fit.

Steps/Code to Reproduce

import numpy as np
import pandas as pd
import sklearn.dummy

cls = sklearn.dummy.DummyRegressor(strategy='mean')
df = pd.DataFrame(data={'A': ['foo', 'bar', 'baz']})
X = df.loc[:, ['A']].values  # X's data is of type object
y = [1, 2, 3]
cls.fit(X, y)
cls.predict(X)

Expected Result

returns [2, 2, 2] and no ValueError at all.

Actual Result

Traceback (most recent call last):
  File "<input>", line 1, in <module>
  File ".../python3.6/site-packages/sklearn/dummy.py", line 468, in predict
    X = check_array(X, accept_sparse=['csr', 'csc', 'coo'])
  File ".../python3.6/site-packages/sklearn/utils/validation.py", line 382, in check_array
    array = np.array(array, dtype=dtype, order=order, copy=copy)
ValueError: could not convert string to float: 'baz'

Versions

Python 3.6.1 | packaged by conda-forge | (default, Mar 23 2017, 21:57:00)
[GCC 4.2.1 Compatible Apple LLVM 6.1.0 (clang-602.0.53)] on darwin
NumPy 1.12.1
SciPy 0.19.0
Scikit-Learn 0.18.1

Proposal

in sklearn.dummy.DummyRegressor, method predict call check_array with kwargs dtype=None and ensure_min_features=0

[jnothman]

I think this may be fixed in 0.19. Please check

…

On 26 September 2017 at 20:00, David Catteeuw ***@***.***> wrote: Description sklearn.dummy.DummyRegressor requires X to be numeric and have at least 1 feature in predict. IMO, this is too strict. Also, it is not done in fit. Steps/Code to Reproduce import numpy as np import pandas as pd import sklearn.dummy cls = sklearn.dummy.DummyRegressor(strategy='mean') df = pd.DataFrame(data={'A': ['foo', 'bar', 'baz']}) X = df.loc[:, ['A']].values # X's data is of type object y = [1, 2, 3] cls.fit(X, y) cls.predict(X) Expected Result returns [2, 2, 2] and no ValueError at all. Actual Result Traceback (most recent call last): File "<input>", line 1, in <module> File ".../python3.6/site-packages/sklearn/dummy.py", line 468, in predict X = check_array(X, accept_sparse=['csr', 'csc', 'coo']) File ".../python3.6/site-packages/sklearn/utils/validation.py", line 382, in check_array array = np.array(array, dtype=dtype, order=order, copy=copy) ValueError: could not convert string to float: 'baz' Versions Python 3.6.1 | packaged by conda-forge | (default, Mar 23 2017, 21:57:00) [GCC 4.2.1 Compatible Apple LLVM 6.1.0 (clang-602.0.53)] on darwin NumPy 1.12.1 SciPy 0.19.0 Scikit-Learn 0.18.1 Proposal in sklearn.dummy.DummyRegressor, method predict call check_array with kwargs dtype=None and ensure_min_features=0 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9832>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEz6zcb6jQBHWtfRYju4kn8b8szX69Lks5smMtJgaJpZM4Pj_QK> .

[dcatteeu]

Good point. I forgot to mention there was a similar issue that is fixed in 0.19: #8916

In 0.19, check_array is called with force_all_finite=False. This means you can now pass data with missing values, but X must still be numeric and have at least 1 feature.

[jnothman]

Okay

[jnothman]

Do we need it to be an array?