如何在 Sklearn 中执行 OneHotEncoding，获取值错误

Question

I just started learning machine learning, when practicing one of the task, I am getting value error, but I followed the same steps as the instructor does.

I am getting value error, please help.

dff

     Country    Name
 0     AUS      Sri
 1     USA      Vignesh
 2     IND      Pechi
 3     USA      Raj

First I performed labelencoding,

X=dff.values
label_encoder=LabelEncoder()
X[:,0]=label_encoder.fit_transform(X[:,0])

out:
X
array([[0, 'Sri'],
       [2, 'Vignesh'],
       [1, 'Pechi'],
       [2, 'Raj']], dtype=object)

then performed One hot encoding for the same X

onehotencoder=OneHotEncoder( categorical_features=[0])
X=onehotencoder.fit_transform(X).toarray()

I am getting the below error:

ValueError                                Traceback (most recent call last)
<ipython-input-472-be8c3472db63> in <module>()
----> 1 X=onehotencoder.fit_transform(X).toarray()

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\preprocessing\data.py in fit_transform(self, X, y)
   1900         """
   1901         return _transform_selected(X, self._fit_transform,
-> 1902                                    self.categorical_features, copy=True)
   1903 
   1904     def _transform(self, X):

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\preprocessing\data.py in _transform_selected(X, transform, selected, copy)
   1695     X : array or sparse matrix, shape=(n_samples, n_features_new)
   1696     """
-> 1697     X = check_array(X, accept_sparse='csc', copy=copy, dtype=FLOAT_DTYPES)
   1698 
   1699     if isinstance(selected, six.string_types) and selected == "all":

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\utils\validation.py in check_array(array, accept_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, ensure_min_samples, ensure_min_features, warn_on_dtype, estimator)
    380                                       force_all_finite)
    381     else:
--> 382         array = np.array(array, dtype=dtype, order=order, copy=copy)
    383 
    384         if ensure_2d:

ValueError: could not convert string to float: 'Raj'

Please edit my question is anything wrong, thanks in advance!

Answer 1

You can go directly to OneHotEncoding now without using the LabelEncoder , and as we move toward version 0.22 many might want to do things this way to avoid warnings and potential errors ( see DOCS and EXAMPLES ).

---

Example code 1 where ALL columns are encoded and where the categories are explicitly specified:

import pandas as pd
import numpy as np
from sklearn.preprocessing import OneHotEncoder

data= [["AUS", "Sri"],["USA","Vignesh"],["IND", "Pechi"],["USA","Raj"]]

df = pd.DataFrame(data, columns=['Country', 'Name'])
X = df.values

countries = np.unique(X[:,0])
names = np.unique(X[:,1])

ohe = OneHotEncoder(categories=[countries, names])
X = ohe.fit_transform(X).toarray()

print (X)

---

Output for code example 1:

[[1. 0. 0. 0. 0. 1. 0.]
 [0. 0. 1. 0. 0. 0. 1.]
 [0. 1. 0. 1. 0. 0. 0.]
 [0. 0. 1. 0. 1. 0. 0.]]

---

Example code 2 showing the 'auto' option for specification of categories:

The first 3 columns encode the country names, the last four the personal names.

import pandas as pd
import numpy as np
from sklearn.preprocessing import OneHotEncoder

data= [["AUS", "Sri"],["USA","Vignesh"],["IND", "Pechi"],["USA","Raj"]]

df = pd.DataFrame(data, columns=['Country', 'Name'])
X = df.values

ohe = OneHotEncoder(categories='auto')
X = ohe.fit_transform(X).toarray()

print (X)

Output for code example 2 (same as for 1):

[[1. 0. 0. 0. 0. 1. 0.]
 [0. 0. 1. 0. 0. 0. 1.]
 [0. 1. 0. 1. 0. 0. 0.]
 [0. 0. 1. 0. 1. 0. 0.]]

---

Example code 3 where only the first column is one hot encoded:

Now, here's the unique part. What if you only need to One Hot Encode a specific column for your data?

( Note: I've left the last column as strings for easier illustration. In reality it makes more sense to do this WHEN the last column was already numerical).

import pandas as pd
import numpy as np
from sklearn.preprocessing import OneHotEncoder

data= [["AUS", "Sri"],["USA","Vignesh"],["IND", "Pechi"],["USA","Raj"]]

df = pd.DataFrame(data, columns=['Country', 'Name'])
X = df.values

countries = np.unique(X[:,0])
names = np.unique(X[:,1])

ohe = OneHotEncoder(categories=[countries]) # specify ONLY unique country names
tmp = ohe.fit_transform(X[:,0].reshape(-1, 1)).toarray()

X = np.append(tmp, names.reshape(-1,1), axis=1)

print (X)

---

Output for code example 3:

[[1.0 0.0 0.0 'Pechi']
 [0.0 0.0 1.0 'Raj']
 [0.0 1.0 0.0 'Sri']
 [0.0 0.0 1.0 'Vignesh']]

Answer 2

Below implementation should work well. Note that the input of onehotencoder fit_transform must not be 1-rank array and also output is sparse and we have used to_array() to expand it.

import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder
from sklearn.preprocessing import OneHotEncoder

data= [["AUS", "Sri"],["USA","Vignesh"],["IND", "Pechi"],["USA","Raj"]]


df = pd.DataFrame(data, columns=['Country', 'Name'])
X = df.values

le = LabelEncoder()
X_num = le.fit_transform(X[:,0]).reshape(-1,1)

ohe = OneHotEncoder()
X_num = ohe.fit_transform(X_num)

print (X_num.toarray())

X[:,0] = X_num

print (X)

Answer 3

An alternative if you do want to encode multiple categorical features is to use a Pipeline with a FeatureUnion and a couple custom Transformers.

First need two transformers - one for selecting a single column and one for making LabelEncoder usable in a Pipeline (The fit_transform method only takes X, it needs to take an optional y to work in a Pipeline).

from sklearn.base import BaseEstimator, TransformerMixin

class SingleColumnSelector(TransformerMixin, BaseEstimator):
    def __init__(self, column):
        self.column = column

    def transform(self, X, y=None):
        return X[:, self.column].reshape(-1, 1)

    def fit(self, X, y=None):
        return self

class PipelineAwareLabelEncoder(TransformerMixin, BaseEstimator):
    def fit(self, X, y=None):
        return self

    def transform(self, X, y=None):
        return LabelEncoder().fit_transform(X).reshape(-1, 1)

Next create a Pipeline (or just a FeatureUnion) which has 2 branches - one for each of the categorical columns. Within each select 1 column, encode the labels and then one hot encode.

import pandas as pd
import numpy as np

from sklearn.preprocessing import LabelEncoder, OneHotEncoder, FunctionTransformer
from sklearn.pipeline import Pipeline, make_pipeline, FeatureUnion

pipeline = Pipeline([(
    'encoded_features',
    FeatureUnion([('countries',
        make_pipeline(
            SingleColumnSelector(0),
            PipelineAwareLabelEncoder(),
            OneHotEncoder()
        )), 
        ('names', make_pipeline(
            SingleColumnSelector(1),
            PipelineAwareLabelEncoder(),
            OneHotEncoder()
        ))
    ]))
])

Finally run your full dataframe through the Pipeline - it will one hot encode each column separately and concatenate at the end.

df = pd.DataFrame([["AUS", "Sri"],["USA","Vignesh"],["IND", "Pechi"],["USA","Raj"]], columns=['Country', 'Name'])
X = df.values
transformed_X = pipeline.fit_transform(X)
print(transformed_X.toarray())

Which returns (first 3 columns are the countries, second 4 are the names)

[[ 1.  0.  0.  0.  0.  1.  0.]
 [ 0.  0.  1.  0.  0.  0.  1.]
 [ 0.  1.  0.  1.  0.  0.  0.]
 [ 0.  0.  1.  0.  1.  0.  0.]]

Answer 4

to make the long story short, if you are looking to dummify your df, use dummy=pd.get_dummies as:

dummy=pd.get_dummies(df['str'])
df=pd.concat([df,dummy], axis=1)
print(Data)