7

I have a dataframe X with integer, float and string columns. I'd like to one-hot encode every column that is of "Object" type, so I'm trying to do this:

encoding_needed = X.select_dtypes(include='object').columns
ohe = preprocessing.OneHotEncoder()
X[encoding_needed] = ohe.fit_transform(X[encoding_needed].astype(str)) #need astype bc I imputed with 0, so some rows have a mix of zeroes and strings.

However, I end up with IndexError: tuple index out of range. I don't quite understand this as per the documentation the encoder expects X: array-like, shape [n_samples, n_features], so I should be OK passing a dataframe. How can I one-hot encode the list of columns specifically marked in encoding_needed?

EDIT:

The data is confidential so I cannot share it and I cannot create a dummy as it has 123 columns as is.

I can provide the following:

X.shape: (40755, 123)
encoding_needed.shape: (81,) and is a subset of columns.

Full stack:

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-90-6b3e9fdb6f91> in <module>()
      1 encoding_needed = X.select_dtypes(include='object').columns
      2 ohe = preprocessing.OneHotEncoder()
----> 3 X[encoding_needed] = ohe.fit_transform(X[encoding_needed].astype(str))

~/anaconda3/envs/python3/lib/python3.6/site-packages/pandas/core/frame.py in __setitem__(self, key, value)
   3365             self._setitem_frame(key, value)
   3366         elif isinstance(key, (Series, np.ndarray, list, Index)):
-> 3367             self._setitem_array(key, value)
   3368         else:
   3369             # set column

~/anaconda3/envs/python3/lib/python3.6/site-packages/pandas/core/frame.py in _setitem_array(self, key, value)
   3393                 indexer = self.loc._convert_to_indexer(key, axis=1)
   3394                 self._check_setitem_copy()
-> 3395                 self.loc._setitem_with_indexer((slice(None), indexer), value)
   3396 
   3397     def _setitem_frame(self, key, value):

~/anaconda3/envs/python3/lib/python3.6/site-packages/pandas/core/indexing.py in _setitem_with_indexer(self, indexer, value)
    592                     # GH 7551
    593                     value = np.array(value, dtype=object)
--> 594                     if len(labels) != value.shape[1]:
    595                         raise ValueError('Must have equal len keys and value '
    596                                          'when setting with an ndarray')

IndexError: tuple index out of range
lte__
  • 7,175
  • 25
  • 74
  • 131

3 Answers3

18
# example data
X = pd.DataFrame({'int':[0,1,2,3],
                   'float':[4.0, 5.0, 6.0, 7.0],
                   'string1':list('abcd'),
                   'string2':list('efgh')})

   int  float string1 string2
0    0    4.0       a       e
1    1    5.0       b       f
2    2    6.0       c       g
3    3    7.0       d       h

Using pandas

With pandas.get_dummies, it will automatically select your object columns and drop these columns while appenind the one-hot-encoded columns:

pd.get_dummies(X)

   int  float  string1_a  string1_b  string1_c  string1_d  string2_e  \
0    0    4.0          1          0          0          0          1   
1    1    5.0          0          1          0          0          0   
2    2    6.0          0          0          1          0          0   
3    3    7.0          0          0          0          1          0   

   string2_f  string2_g  string2_h  
0          0          0          0  
1          1          0          0  
2          0          1          0  
3          0          0          1

Using sklearn

Here we have to specify that we only need the object columns:

from sklearn.preprocessing import OneHotEncoder

ohe = OneHotEncoder()

X_object = X.select_dtypes('object')
ohe.fit(X_object)

codes = ohe.transform(X_object).toarray()
feature_names = ohe.get_feature_names(['string1', 'string2'])

X = pd.concat([df.select_dtypes(exclude='object'), 
               pd.DataFrame(codes,columns=feature_names).astype(int)], axis=1)

   int  float  string1_a  string1_b  string1_c  string1_d  string2_e  \
0    0    4.0          1          0          0          0          1   
1    1    5.0          0          1          0          0          0   
2    2    6.0          0          0          1          0          0   
3    3    7.0          0          0          0          1          0   

   string2_f  string2_g  string2_h  
0          0          0          0  
1          1          0          0  
2          0          1          0  
3          0          0          1
Erfan
  • 40,971
  • 8
  • 66
  • 78
  • While doing ohe.fit(df["sales"]) I am getting error as ValueError: Expected 2D array, got 1D array instead: array=['ab' 'vg' 'ab' 'iu' 'ab' 'vg' 'iu']. Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample. – Shashank Feb 19 '22 at 16:54
1

Without seeing your data, I'm having a hard time finding your error. You could try the get_dummies method from pandas? 

pd.get_dummies(X[encoding_needed])
cmxu
  • 954
  • 5
  • 13
  • `X[encoding_needed] = pd.get_dummies(X[encoding_needed])` results in a `ValueError: Columns must be same length as key`. – lte__ Feb 10 '20 at 16:19
1

In case of 'get_feature_names not found' for OneHotEncoder, the following might be more feasible:

import pandas as pd
columns_encode=['string1','string2']
encoder = OneHotEncoder()
df_X_enumeric=X.copy()

for col in columns_encode:
  onehot = encoder.fit_transform(df_X_enumeric[[col]])
  feature_names = encoder.categories_[0]
  onehot_df = pd.DataFrame(onehot.toarray(), columns=feature_names)
  df_X_enumeric= pd.concat([df_X_enumeric, onehot_df], axis=1)


df_X_enumeric.drop(columns_encode, axis=1, inplace=True)

,oneHot vs dummies is also helpful.

msbeigi
  • 189
  • 1
  • 5