Standardizing Features

June 01, 2019

Machine learning models can be sensitive to the magnitude of their inputs. If one or several features in your data are of a larger magnitude than the rest, they may have undue influence over the model’s output. Scaling your features to have similar magnitude can help models to better generalize and converge faster.

Imports

import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import MinMaxScaler

Generate Data

# create two similar features with different scales
df = pd.DataFrame({"X1": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
	"X2": [0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]})
print(df)

Method 1: StandardScaler

This method scales all features to have a mean of 0 and variance of 1. This is my scaler of choice in most cases because it standardizes both magnitude and variance.

Fit the Scaler and Apply its Transformation

# save column names before transforming because standardscaler throws them out
cols = df.columns
df_scaled = pd.DataFrame(StandardScaler().fit_transform(df))
df_scaled.columns = cols
print(df_scaled)

Method 2: Min-Max Scaler

This method scales all features to have the same minimum and maximum values. Common min/max values for this are 0 and 1.

Fit the Scaler and Apply its Transformation

# save column names before transforming because standardscaler throws them out
cols = df.columns
df_scaled = pd.DataFrame(MinMaxScaler(feature_range=(0, 1)).fit_transform(df))
df_scaled.columns = cols
print(df_scaled)