Vectorization vs apply() vs iterrows(): Performance Benchmarks in Pandas

Question

Given a DataFrame with 1 million rows containing `quantity` and `unit_price`, compute `total_revenue = quantity * unit_price`. Compare the performance of `iterrows()`, `.apply(axis=1)`, and direct vectorization.

Accepted Answer

Optimal Solution:
import pandas as pd
import numpy as np

# Method 1: Vectorized (C-Level SIMD Execution) - FASTEST
df['total_revenue'] = df['quantity'] * df['unit_price']

# Method 2: numpy.where for conditional vectorization
df['discounted_rev'] = np.where(df['quantity'] > 10, df['total_revenue'] * 0.9, df['total_revenue'])

Technical Explanation:
Direct vectorization operates on contiguous C memory buffers using SIMD instructions, bypassing the Python GIL and object allocation overhead. `iterrows()` creates a Python Series object on every single row.

method	runtime
df['q'] * df['p'] (Vectorized)	3 ms (Fastest)
df.apply(axis=1)	1,200 ms
for row in df.iterrows()	28,000 ms (Slowest)