[BUG] Expected array-like (array or non-string sequence), got 0 0.0

[sbushmanov]

Describe the bug
Trying to estimate accuracy of pandas.core.series.Seriesvs cudf.core.series.Series
Getting error message:
ValueError: Expected array-like (array or non-string sequence), got 0 0.0

Steps/Code to reproduce bug

from cuml.ensemble import RandomForestClassifier as curfc
from sklearn.metrics import accuracy_score
cuml_model = curfc(n_estimators=40,
                   max_depth=16,
                   max_features=1.0,
                   seed=10)

cuml_model.fit(X_cudf_train, y_cudf_train)
fil_preds_orig = cuml_model.predict(X_cudf_test)
fil_acc_orig = accuracy_score(y_test,fil_preds_orig)
ValueError                                Traceback (most recent call last)
<ipython-input-45-3e552f306f92> in <module>
      1 # %%time
      2 fil_preds_orig = cuml_model.predict(X_cudf_test)
----> 3 fil_acc_orig = accuracy_score(y_test,fil_preds_orig)
      4 fil_acc_orig

~/anaconda3/lib/python3.7/site-packages/sklearn/metrics/_classification.py in accuracy_score(y_true, y_pred, normalize, sample_weight)
    183 
    184     # Compute accuracy for each possible representation
--> 185     y_type, y_true, y_pred = _check_targets(y_true, y_pred)
    186     check_consistent_length(y_true, y_pred, sample_weight)
    187     if y_type.startswith('multilabel'):

~/anaconda3/lib/python3.7/site-packages/sklearn/metrics/_classification.py in _check_targets(y_true, y_pred)
     80     check_consistent_length(y_true, y_pred)
     81     type_true = type_of_target(y_true)
---> 82     type_pred = type_of_target(y_pred)
     83 
     84     y_type = {type_true, type_pred}

~/anaconda3/lib/python3.7/site-packages/sklearn/utils/multiclass.py in type_of_target(y)
    239     if not valid:
    240         raise ValueError('Expected array-like (array or non-string sequence), '
--> 241                          'got %r' % y)
    242 
    243     sparse_pandas = (y.__class__.__name__ in ['SparseSeries', 'SparseArray'])

ValueError: Expected array-like (array or non-string sequence), got 0        0.0
1        1.0
2        1.0
3        1.0
4        0.0
        ... 
26210    1.0
26211    0.0
26212    1.0
26213    1.0
26214    0.0
Length: 26215, dtype: float32

Expected behavior
Estimation of accuracy is expected

Environment details (please complete the following information):

• Environment location: Bare-metal
• Linux Distro/Architecture: [Ubuntu 16.04 amd64]
• GPU Model/Driver: NVIDIA-SMI 440.59 Driver Version: 440.59
• CUDA Version: 10.2
• Method of cuDF & cuML install: conda
  Name Version Build Channel
  cuml 0.13.0a200317 cuda10.2_py37_1494 rapidsai-nightly
  libcuml 0.13.0a200317 cuda10.2_1494 rapidsai-nightly
  libcumlprims 0.13.0a200313 cuda10.2_11 rapidsai-nightly

The error happens after updating for the latest nightly build. It used to run fine.

[dantegd]

@sbushmanov this seems an issue with Scikit-learn expecting an array as opposed to a cuDF Series, to be able to use that particular accuracy method you can convert the cuDF Series to a NumPy array like:

fil_acc_orig = accuracy_score(y_test, fil_preds_orig.to_array())

alternatively you can also use the accuracy from cuML itself which does accept cuDF Series (or NumPy, CuPy, Numba arrays among other formats): https://rapidsai.github.io/projects/cuml/en/0.12.0/api.html?highlight=accuracy#cuml.metrics.accuracy.accuracy_score

[sbushmanov]

@dantegd Thanks for answering!

The code snippet used to work in previous releases (and it is still present in the official notebooks).

Did they change the policy on how they supply the arrays to sklearn?

[divyegala]

Closing this via rapidsai/cudf#4864