LabelEncoder throws an error when it's used in a Pipeline or in a ColumnTransform

[hal-314]

Description

fit and fit_transform methods in LabelEncoder don't follow the standard scikit-lean convention for these methods: fit(X[, y]) and fit_transform(X[, y]). The fit and fit_transform method in the LabelEncoder only accepts one argument: fit(y) and fit_transform(y).

Therefore, LabelEncoder couldn't be used inside a Pipeline or a ColumnTransform. I suspect that there are a bunch of other classes in which it doesn't work (GridSearchCV, ...) but I haven't tested it.

In contrast, fit and fit_transform methods in OneHotEncoder and OrdinalEncoder follows the standard scikit-learn signature.

See reference:
LabelEncoder: https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.LabelEncoder.html
OneHotEnconder: https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html#sklearn.preprocessing.OneHotEncoder
OrdinalEncoder:https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OrdinalEncoder.html#sklearn.preprocessing.OrdinalEncoder

Steps/Code to Reproduce

Example:

import pandas as pd

from sklearn.preprocessing import LabelEncoder
from sklearn.pipeline import make_pipeline
from sklearn.compose import ColumnTransformer
import sklearn.tree as tree

X = pd.DataFrame(
    {'city': ['London', 'London', 'Paris', 'Sallisaw'],
     'title': ["His Last Bow", "How Watson Learned the Trick",
               "A Moveable Feast", "The Grapes of Wrath"],
     'expert_rating': [5, 3, 4, 5],
     'user_rating': [4, 5, 4, 3]})

column_trans = ColumnTransformer(
    [('title_bow', LabelEncoder(), 'title')],
    remainder='drop').fit(X)

pipe = make_pipeline(LabelEncoder(), tree.DecisionTreeClassifier()).fit(X)

Expected Results

No error is thrown.

Actual Results

The same error in both cases:
TypeError: fit_transform() takes 2 positional arguments but 3 were given.

Versions

System:
python: 3.6.6 |Anaconda, Inc.| (default, Oct 9 2018, 12:34:16) [GCC 7.3.0]
executable: /home/twins/anaconda3/envs/pytorch/bin/python
machine: Linux-4.8.0-56-generic-x86_64-with-debian-stretch-sid

BLAS:
macros: SCIPY_MKL_H=None, HAVE_CBLAS=None
lib_dirs: /home/twins/anaconda3/envs/pytorch/lib
cblas_libs: mkl_rt, pthread

Python deps:
pip: 18.1
setuptools: 40.6.2
sklearn: 0.20.1
numpy: 1.15.4
scipy: 1.1.0
Cython: None
pandas: 0.23.4

Thanks for the amazing job you do !

[TomDLT]

Yes indeed, from the user guide:

These are transformers that are not intended to be used on features, only on supervised learning targets.

To encode features, you need to use OneHotEncoder or OrdinalEncoder.

[qinhanmin2014]

duplicate of #3956
Please use the latest OneHotEncoder (in scikit-learn >= 0.20)

[gmichaelson]

The problem is that OneHotEncoder is broken in >0.20... it fails if you pass it features with string values...

[maggiex]

The problem is that OneHotEncoder is broken in >0.20... it fails if you pass it features with string values...

yes.. any suggestions on how to resolve this?

[qinhanmin2014]

The problem is that OneHotEncoder is broken in >0.20... it fails if you pass it features with string values...

Why? See the example section in our doc. https://scikit-learn.org/dev/modules/generated/sklearn.preprocessing.OneHotEncoder.html

[kuabhish]

Yes indeed, from the user guide:

These are transformers that are not intended to be used on features, only on supervised learning targets.

To encode features, you need to use OneHotEncoder or OrdinalEncoder.

What if I want to Labelencode the input feature?
With Labelencode there is also problem with unseen category (for e.x in the test set)
What should I do then?

[trialzuki]

may be cos LabelEncoder wont pass multi_future in one step, but OneHotEncoder can do this ?

check this:
OneHotEncoder().fit(df[multi_cols]) - pass OK
vs
LabelEncoder().fit(df[multi_cols]) - get error

[kuabhish]

I have written a labelencoder which can take care of unknown values and multiple columns too and can be inserted into a pipeline with some tweaks.

Code:
`
class My_LabelEncoder(BaseEstimator, TransformerMixin):

def fit( self , df ,df_y  ):
  maps_={}
  for col in df:
    y = df[col]
    uni = np.unique(y)
    map_ = {}
    for c in uni:
        map_[c] = len(map_)
    maps_[col] = map_
  self.maps_ = maps_
  return self


def transform(self , df):
  ndf = df.copy()
  for col in df:
    ny = []
    map_= self.maps_[col]
    for c in np.array(df[col]):
      if c in self.maps_[col]:
        ny.append(self.maps_[col][c])
      else:
        ny.append(-1)
    ndf[col] = ny
  return ndf`

[jinsel]

Here is an example for Multiple Column convert into int

from sklearn.preprocessing import LabelEncoder
column= ['ethnicity','gender','hospital_admit_source','icu_stay_type','icu_type','icu_admit_source','icu_id','apache_3j_bodysystem','apache_2_bodysystem']
def cat_to_int(column):
for col in column:
encoder = LabelEncoder()
str_to_int = df_training1[col]
df_training1[col]=encoder.fit_transform(str_to_int)
cat_to_int (column)

[jnothman]

Please don't do that. Please use OrdinalEncoder, perhaps together with ColumnTransformer.

[Vikrant-Deshmukh]

Traceback (most recent call last):
File "/Users/vikrant/Downloads/Car/catthatcodes/app/app.py", line 7, in
from pipeline import pipeline
File "/Users/vikrant/Downloads/Car/catthatcodes/app/pipeline/pipeline.py", line 8, in
graph = tf.get_default_graph()
AttributeError: module 'tensorflow' has no attribute 'get_default_graph'

#Here is the Code
import json
import urllib.request
import numpy as np
import pickle as pk

import tensorflow as tf
global graph, model
graph = tf.get_default_graph()

from tensorflow.keras.models import load_model
from tensorflow.keras.applications.vgg16 import VGG16
from tensorflow.keras.applications.imagenet_utils import preprocess_input
from tensorflow.keras.preprocessing.image import img_to_array, load_img
import keras.utils.data_utils