Lære Scaling and Normalizing Numerical Features

Sveip for å vise menyen

Numeric columns in the flights dataset have very different ranges – DISTANCE spans hundreds to thousands, while DEPARTURE_DELAY is typically under 100. Many ML algorithms are sensitive to scale: a column with large values will dominate the model unless you normalize the features first.

StandardScaler

StandardScaler standardizes each feature to have mean 0 and standard deviation 1. It requires the input to be a vector column, so you first assemble the numeric columns:


              123456789101112131415161718192021222324252627282930
            
import urllib.request
from pyspark.sql import SparkSession
from pyspark.ml.feature import VectorAssembler, StandardScaler

urllib.request.urlretrieve(
    "https://staging-content-media-cdn.codefinity.com/courses/aa80ac56-0d50-49e8-9231-2c2374cd3e9d/flights.csv",
    "flights.csv"
)

spark = SparkSession.builder \
    .appName("Scaling") \
    .master("local[*]") \
    .getOrCreate()

flights_df = spark.read.csv("flights.csv", header=True, inferSchema=True) \
    .fillna(0, subset=["DEPARTURE_DELAY", "ARRIVAL_DELAY", "DISTANCE"])

# Assembling numeric columns into a single vector
assembler = VectorAssembler(
    inputCols=["DEPARTURE_DELAY", "ARRIVAL_DELAY", "DISTANCE"],
    outputCol="FEATURES_RAW"
)
flights_df = assembler.transform(flights_df)

# Standardizing to mean=0, std=1
scaler = StandardScaler(inputCol="FEATURES_RAW", outputCol="FEATURES_SCALED", withMean=True, withStd=True)
scaler_model = scaler.fit(flights_df)
flights_df = scaler_model.transform(flights_df)

flights_df.select("FEATURES_RAW", "FEATURES_SCALED").show(3, truncate=False)

MinMaxScaler

MinMaxScaler rescales each feature to a fixed range, by default [0, 1]:


              123456
            
from pyspark.ml.feature import MinMaxScaler

min_max_scaler = MinMaxScaler(inputCol="FEATURES_RAW", outputCol="FEATURES_MINMAX")
flights_df = min_max_scaler.fit(flights_df).transform(flights_df)

flights_df.select("FEATURES_RAW", "FEATURES_MINMAX").show(3, truncate=False)

Use StandardScaler when your algorithm assumes normally distributed features (e.g. linear regression, SVMs). Use MinMaxScaler when you need values bounded in a specific range (e.g. neural networks).

Alt var klart?

Takk for tilbakemeldingene dine!

Seksjon 1. Kapittel 3

Spør AI

Spør om hva du vil, eller prøv ett av de foreslåtte spørsmålene for å starte chatten vår

Seksjon 1. Kapittel 3