Those features which contain constant values (i.e. only one value for all the outputs or target values) in the dataset are known as Constant Features. These features don’t provide any information to the target feature. These are redundant data available in the dataset. Presence of this feature has no effect on the target, so it is good to remove these features from the dataset. This process of removing redundant features and keeping only the necessary features in the dataset comes under the filter method of Feature Selection Methods.
Now Let’s see how we can remove constant features in Python.
Consider the self created dataset for the article:
| Portal | Article’s_category | Views |
|---|---|---|
| Lazyroar | Python | 545 |
| Lazyroar | Data Science | 1505 |
| Lazyroar | Data Science | 1157 |
| Lazyroar | Data Science | 2541 |
| Lazyroar | Mathematics | 5726 |
| Lazyroar | Python | 3125 |
| Lazyroar | Data Science | 3131 |
| Lazyroar | Mathematics | 6525 |
| Lazyroar | Mathematics | 15000 |
Code: Create DataFrame of the above data
# Import pandas to create DataFrame import pandas as pd # Make DataFrame of the given data data = pd.DataFrame({"Portal":['Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar', 'Lazyroar'], "Article's_category":['Python', 'Data Science', 'Data Science', 'Data Science', 'Mathematics', 'Python', 'Data Science', 'Mathematics', 'Mathematics'], "Views":[545, 1505, 1157, 2541, 5726, 3125, 3131, 6525, 15000]}) |
Code: Convert the categorical data to numerical data
# import ordinal encoder from sklearn from sklearn.preprocessing import OrdinalEncoder ord_enc = OrdinalEncoder() # Transform the data data[["Portal","Article's_category"]] = ord_enc.fit_transform(data[["Portal","Article's_category"]]) |
Code: Fit the data to VarianceThreshold.
# import VarianceThreshold from sklearn.feature_selection import VarianceThreshold var_threshold = VarianceThreshold(threshold=0) # threshold = 0 for constant # fit the data var_threshold.fit(data) # We can check the variance of different features as print(var_threshold.variances_) |
Output: Variance of different features:
[0.00000000e+00 6.17283951e-01 1.76746269e+07]
Code: Transform the data
print(var_threshold.transform(data)) print('*' * 10,"Separator",'*' * 10) # shapes of data before transformed and after transformed print("Earlier shape of data: ", data.shape) print("Shape after transformation: ", var_threshold.transform(data).shape) |
Output:
[[2.000e+00 5.450e+02] [0.000e+00 1.505e+03] [0.000e+00 1.157e+03] [0.000e+00 2.541e+03] [1.000e+00 5.726e+03] [2.000e+00 3.125e+03] [0.000e+00 3.131e+03] [1.000e+00 6.525e+03] [1.000e+00 1.500e+04]] ********** Separator ********** Earlier shape of data: (9, 3) Shape after transformation: (9, 2)
As you can observe earlier we had 9 observations with 3 features.
After transformation we have 9 observations with 2 features. We can clearly observe that the removed feature is ‘Portal’.
