CODEWITHSUNDEEP
pythonpandasdataframescikit-learnfeature-extraction
Noor
Noor

Reputation: 136

Feature Hashing on multiple categorical features(columns)

I would like to hash feature ‘Genre’ into 6 columns and separately feature ‘Publisher’ into another six columns. I want something like below:

      Genre      Publisher  0    1    2    3    4    5      0    1    2    3    4    5 
0     Platform  Nintendo  0.0  2.0  2.0 -1.0  1.0  0.0    0.0  2.0  2.0 -1.0  1.0  0.0
1       Racing      Noir -1.0  0.0  0.0  0.0  0.0 -1.0   -1.0  0.0  0.0  0.0  0.0 -1.0
2       Sports     Laura -2.0  2.0  0.0 -2.0  0.0  0.0   -2.0  2.0  0.0 -2.0  0.0  0.0
3  Roleplaying      John -2.0  2.0  2.0  0.0  1.0  0.0   -2.0  2.0  2.0  0.0  1.0  0.0
4       Puzzle      John  0.0  1.0  1.0 -2.0  1.0 -1.0    0.0  1.0  1.0 -2.0  1.0 -1.0
5     Platform      Noir  0.0  2.0  2.0 -1.0  1.0  0.0    0.0  2.0  2.0 -1.0  1.0  0.0

The following code does what I want to do 

import pandas as pd
d = {'Genre': ['Platform', 'Racing','Sports','Roleplaying','Puzzle','Platform'], 'Publisher': ['Nintendo', 'Noir','Laura','John','John','Noir']}
df = pd.DataFrame(data=d)
from sklearn.feature_extraction import FeatureHasher
fh1 = FeatureHasher(n_features=6, input_type='string')
fh2 = FeatureHasher(n_features=6, input_type='string')
hashed_features1 = fh.fit_transform(df['Genre'])
hashed_features2 = fh.fit_transform(df['Publisher'])
hashed_features1 = hashed_features1.toarray()
hashed_features2 = hashed_features2.toarray()
pd.concat([df[['Genre', 'Publisher']], pd.DataFrame(hashed_features1),pd.DataFrame(hashed_features2)],
          axis=1)

This works for the above two feature but If I have lets say 40 categorical features then this approach would be tedious. Is there any other way to do?

Upvotes: 11

Views: 5098

Answers (2)

Ashok K Harnal
Ashok K Harnal

Reputation: 1221

Even though, I am late here, from the examples I have seen on Kaggle , FeatureHashing is performed at once for multiple columns (ie on a DataFrame) rather than for individual columns and concatenating the sparse matrices. See Notebooks on Kaggle, here and here. I have also used both ways of performing feature hashing on this data, ie:

a. Hash individual categorical columns and concatenate the results
b. Hash all categorical columns of a DataFrame at once

Logistic Regression classifier gave significantly better results when approach (b) was followed rather than approach (a).

Upvotes: 2

Jan K
Jan K

Reputation: 4150

Hashing (Update)

Assuming that new categories might show up in some of the features, hashing is the way to go. Just 2 notes:

  • Be aware of the possibility of collision and adjust the number of features accordingly
  • In your case, you want to hash each feature separately

One Hot Vector

In case the number of categories for each feature is fixed and not too large, use one hot encoding.

I would recommend using either of the two:

  1. sklearn.preprocessing.OneHotEncoder
  2. pandas.get_dummies

Example

import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.feature_extraction import FeatureHasher
from sklearn.preprocessing import OneHotEncoder

df = pd.DataFrame({'feature_1': ['A', 'G', 'T', 'A'],
                   'feature_2': ['cat', 'dog', 'elephant', 'zebra']})

# Approach 0 (Hashing per feature)
n_orig_features = df.shape[1]
hash_vector_size = 6
ct = ColumnTransformer([(f't_{i}', FeatureHasher(n_features=hash_vector_size, 
                        input_type='string'), i) for i in range(n_orig_features)])

res_0 = ct.fit_transform(df)  # res_0.shape[1] = n_orig_features * hash_vector_size

# Approach 1 (OHV)
res_1 = pd.get_dummies(df)

# Approach 2 (OHV)
res_2 = OneHotEncoder(sparse=False).fit_transform(df)

res_0 : 

array([[ 0.,  0.,  0.,  0.,  1.,  0.,  0.,  0.,  1., -1.,  0., -1.],
       [ 0.,  0.,  0.,  1.,  0.,  0.,  0.,  2., -1.,  0.,  0.,  0.],
       [ 0., -1.,  0.,  0.,  0.,  0., -2.,  2.,  2., -1.,  0., -1.],
       [ 0.,  0.,  0.,  0.,  1.,  0.,  0.,  2.,  1., -1.,  0., -1.]])

res_1 : 

   feature_1_A  feature_1_G  feature_1_T  feature_2_cat  feature_2_dog  feature_2_elephant  feature_2_zebra
0            1            0            0              1              0                   0                0
1            0            1            0              0              1                   0                0
2            0            0            1              0              0                   1                0
3            1            0            0              0              0                   0                1

res_2 : 

array([[1., 0., 0., 1., 0., 0., 0.],
       [0., 1., 0., 0., 1., 0., 0.],
       [0., 0., 1., 0., 0., 1., 0.],
       [1., 0., 0., 0., 0., 0., 1.]])

Upvotes: 7

Related Questions