classes should include all valid labels that can be in y

Question

I wanted to pass a matrix of weights of the classes of a dataset to a neural network.

from sklearn.utils import class_weight
class_weights = class_weight.compute_class_weight('balanced',
                                                 np.unique(y_train),
                                                 y_train)

However I get the following error :

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-93-9452aecf4030> in <module>
      2 class_weights = class_weight.compute_class_weight('balanced',
      3                                                  np.unique(y_train),
----> 4                                                  y_train)

~\AppData\Roaming\Python\Python36\site-packages\sklearn\utils\class_weight.py in compute_class_weight(class_weight, classes, y)
     39 
     40     if set(y) - set(classes):
---> 41         raise ValueError("classes should include all valid labels that can "
     42                          "be in y")
     43     if class_weight is None or len(class_weight) == 0:

ValueError: classes should include all valid labels that can be in y

I don't understand, here is part of my y_train dataset:

        grade_A  grade_B  grade_C  grade_D  grade_E  grade_F  grade_G
689526        0        1        0        0        0        0        0
523913        1        0        0        0        0        0        0
266122        0        0        1        0        0        0        0
362552        0        0        0        1        0        0        0

classes [A,B,C,D,E,F] include all valid labels that can be put in y !

Update

I tried to make use .values on the dataframe:

from sklearn.utils import class_weight
class_weights = class_weight.compute_class_weight('balanced',
                                                 np.unique(y_train.values),
                                                 y_train.values)

However it returned:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-25-c2342f04abd9> in <module>
      2 class_weights = class_weight.compute_class_weight('balanced',
      3                                                  np.unique(y_train.values),
----> 4                                                  y_train.values)

~\AppData\Roaming\Python\Python36\site-packages\sklearn\utils\class_weight.py in compute_class_weight(class_weight, classes, y)
     38     from ..preprocessing import LabelEncoder
     39 
---> 40     if set(y) - set(classes):
     41         raise ValueError("classes should include all valid labels that can "
     42                          "be in y")

TypeError: unhashable type: 'numpy.ndarray

If I type print(type(y_train)) I get the following answer:

<class 'pandas.core.frame.DataFrame'>

Answer 1

from sklearn.utils import class_weight
import pandas as pd
import numpy as np

y_train = pd.DataFrame(['dog', 'dog', 'dog', 'dog', 'dog',
                        'cat', 'cat', 'cat', 'bat', 'bat'])

weights = class_weight.compute_class_weight(
    class_weight='balanced',
    classes=np.unique(y_train),
    y=y_train.values.reshape(-1)
)

print(weights)

>> [1.66666667 1.11111111 0.66666667]

Answer 2

Based on the documentaion:

sklearn.utils.class_weight.compute_class_weight(class_weight, classes, y)

classes : ndarray

    Array of the classes occurring in the data, as given by np.unique(y_org) with y_org the original class labels.
y : array-like, shape (n_samples,)

    Array of original class labels per sample;

if you run type(y_train), what are you getting back?

You can convert your dataframe to an array (Convert pandas dataframe to NumPy array):

ytrain = y_train.values