学ぶ Scaling and Normalization | Data Transformation Techniques

メニューを表示するにはスワイプしてください

Numerical features in your data often have very different scales, which can hurt the performance of machine learning algorithms—especially those using distance calculations or assuming normal distributions. Scaling ensures all features contribute equally to model training.

The two main scaling techniques are:

Normalization: rescales features to a fixed range, usually between 0 and 1;
Standardization: transforms features to have a mean of 0 and a standard deviation of 1.

Each method changes your data's range in a different way and is best suited to specific scenarios.


              1234567891011121314151617181920212223242526272829
            
import pandas as pd
from sklearn.preprocessing import StandardScaler, MinMaxScaler

# Load Titanic dataset from seaborn
import seaborn as sns
titanic = sns.load_dataset('titanic')

# Select numerical features for scaling
features = ['age', 'fare', 'sibsp', 'parch']
df = titanic[features].dropna()

# Standardization
scaler_standard = StandardScaler()
df_standardized = pd.DataFrame(
    scaler_standard.fit_transform(df),
    columns=df.columns
)

# Normalization
scaler_minmax = MinMaxScaler()
df_normalized = pd.DataFrame(
    scaler_minmax.fit_transform(df),
    columns=df.columns
)

print("Standardized Data (first 5 rows):")
print(df_standardized.head())
print("\nNormalized Data (first 5 rows):")
print(df_normalized.head())

When to Use Each Scaling Method

Standardization is best when your data follows a Gaussian (normal) distribution, or when algorithms expect centered data, such as linear regression, logistic regression, or k-means clustering.

Normalization is preferred when you want all features to have the same scale, especially for algorithms that use distance metrics, like k-nearest neighbors or neural networks.

すべて明確でしたか？

フィードバックありがとうございます！

セクション 2. 章 1

AIに質問する

何でも質問するか、提案された質問の1つを試してチャットを始めてください

セクション 2. 章 1