Python Performance

Performance profiling, benchmarking, and optimization for Python.

Core Principle

Profile before optimizing - Use profiling tools to identify real bottlenecks. Premature optimization wastes time.

Profiling Tools Decision Matrix

Basic Profiling

cProfile - Function-level

import cProfile
import pstats

profiler = cProfile.Profile()
profiler.enable()

result = expensive_function()

profiler.disable()
stats = pstats.Stats(profiler)
stats.sort_stats('cumulative')
stats.print_stats(20)

line_profiler - Line-level

@profile
def slow_function():
    results = []
    for i in range(10000):
        results.append(i ** 2)
    return results

# Run: kernprof -l -v script.py

See profiling-workflow.md for:

• Complete profiling workflow
• Interpreting profiler output

Optimization Strategies

Algorithm Optimization (Biggest Impact)

# BAD - O(n²)
def find_duplicates_slow(items):
    for i, item in enumerate(items):
        for j, other in enumerate(items[i+1:]):
            if item == other:
                return True

# GOOD - O(n)
def find_duplicates_fast(items):
    return len(items) != len(set(items))

Data Structure Choice

# Use set for membership testing
allowed_set = {1, 2, 3, 4, 5}  # O(1) lookup
if x in allowed_set:
    pass

See optimization-strategies.md for:

• Function call overhead
• String operations
• Dictionary optimizations

NumPy for Numerical Computing

import numpy as np

# BAD - Pure Python loop
result = [x**2 + 2*x + 1 for x in data]

# GOOD - NumPy vectorization (10-100x faster)
arr = np.array(data)
result = arr**2 + 2*arr + 1

See numpy-optimization.md for:

• Broadcasting
• Avoiding loops with vectorization

Numba for JIT Compilation

from numba import jit

@jit(nopython=True)
def monte_carlo_pi_fast(n):
    inside = 0
    for i in range(n):
        x = np.random.random()
        y = np.random.random()
        if x**2 + y**2 <= 1:
            inside += 1
    return 4 * inside / n

See numba-patterns.md for:

• Type signatures
• Parallel execution

Multiprocessing for CPU-Bound Work

from multiprocessing import Pool

def process_parallel(datasets):
    with Pool() as pool:
        return pool.map(cpu_intensive_task, datasets)

See parallel-processing.md for:

• Process vs thread pools
• Shared memory

Performance Anti-Patterns

See performance-anti-patterns.md for examples.

source: Python performance docs