Epochly Examples: Intermediate: Level Selection Guide

import epochly
import numpy as np
@epochly.optimize(level=0)
def baseline_computation(n):
    """Track metrics without optimization"""
    arr = np.random.rand(n)
    result = np.sum(arr ** 2 + np.sin(arr))
    return result
# Run computation
result = baseline_computation(1_000_000)
# Get metrics
metrics = epochly.get_metrics()
print(f"Function calls: {metrics['total_calls']}")
print(f"Total time: {metrics['total_time']:.3f}s")
print(f"Average time: {metrics['avg_time']:.3f}s")

import epochly
import requests
@epochly.optimize(level=1)
def fetch_urls(urls):
    """Fetch multiple URLs concurrently"""
    results = []
    for url in urls:
        response = requests.get(url)
        results.append(response.text)
    return results
# Fetch multiple URLs in parallel
urls = [
    'https://api.example.com/data1',
    'https://api.example.com/data2',
    'https://api.example.com/data3',
]
data = fetch_urls(urls)
print(f"Fetched {len(data)} URLs")

import epochly
import json
@epochly.optimize(level=1)
def load_json_files(file_paths):
    """Load multiple JSON files concurrently"""
    data = []
    for path in file_paths:
        with open(path, 'r') as f:
            data.append(json.load(f))
    return data
files = [f'data_{i}.json' for i in range(100)]
all_data = load_json_files(files)

import epochly
import numpy as np
@epochly.optimize(level=2)
def monte_carlo_pi(n_samples):
    """Estimate Pi using Monte Carlo method with JIT"""
    inside_circle = 0
    
    for _ in range(n_samples):
        x = np.random.random()
        y = np.random.random()
        
        if x**2 + y**2 <= 1:
            inside_circle += 1
    
    return 4 * inside_circle / n_samples
# JIT compiles on first call, fast on subsequent calls
pi_estimate = monte_carlo_pi(10_000_000)
print(f"Pi estimate: {pi_estimate:.6f}")

import epochly
@epochly.optimize(level=2)
def fibonacci_sequence(n):
    """Generate Fibonacci sequence with JIT"""
    if n <= 1:
        return n
    
    a, b = 0, 1
    for _ in range(n - 1):
        a, b = b, a + b
    
    return b
# First call compiles, subsequent calls are fast
result = fibonacci_sequence(1000)

import epochly
import numpy as np
@epochly.optimize(level=3)
def parallel_processing(data_list):
    """Process list items in parallel across cores"""
    results = [process_item(item) for item in data_list]
    return results
def process_item(data):
    """CPU-intensive per-item processing"""
    return np.sum(data ** 2 + np.sin(data) * np.cos(data))
# Generate data
data_list = [np.random.rand(100_000) for _ in range(100)]
# Process in parallel across all cores
results = parallel_processing(data_list)
print(f"Processed {len(results)} items")

import epochly
import pandas as pd
import numpy as np
@epochly.optimize(level=3)
def parallel_groupby(df):
    """Parallel GroupBy aggregation"""
    result = df.groupby('category').agg({
        'value': ['sum', 'mean', 'std'],
        'quantity': ['min', 'max', 'count']
    })
    return result
# Large DataFrame
df = pd.DataFrame({
    'category': np.random.choice(['A', 'B', 'C'], 10_000_000),
    'value': np.random.rand(10_000_000),
    'quantity': np.random.randint(1, 100, 10_000_000)
})
result = parallel_groupby(df)

import epochly
import numpy as np
@epochly.optimize(level=4)
def gpu_matrix_operations(n):
    """Massive matrix operations on GPU"""
    # Create large arrays
    A = np.random.rand(n, n)
    B = np.random.rand(n, n)
    
    # Operations run on GPU
    C = np.dot(A, B)
    D = np.sin(C) + np.cos(C)
    
    return np.sum(D)
# Process 10,000 x 10,000 matrices on GPU
result = gpu_matrix_operations(10_000)
print(f"Result: {result}")

import epochly
import numpy as np
@epochly.optimize(level=4)
def process_images(images):
    """Process batch of images on GPU"""
    # Normalize
    normalized = images / 255.0
    
    # Apply transformations
    transformed = normalized ** 2 + np.sin(normalized)
    
    # Reduce
    features = np.mean(transformed, axis=(1, 2))
    
    return features
# Batch of 1000 images (256x256 RGB)
images = np.random.randint(0, 256, (1000, 256, 256, 3), dtype=np.uint8)
features = process_images(images)

import epochly
@epochly.optimize()  # No level specified
def smart_function(data):
    """Epochly auto-selects the best level"""
    return process(data)
# Epochly analyzes the workload and chooses:
# - Level 1 for I/O operations
# - Level 2 for numerical loops
# - Level 3 for parallel operations
# - Level 4 for large arrays (if GPU available)

import epochly
import numpy as np
@epochly.optimize()  # Auto-select
def auto_optimized(data):
    """Automatically optimized based on workload"""
    # If data is large array -> Level 4
    # If data has loops -> Level 2
    # If data is I/O -> Level 1
    return np.sum(data ** 2)
# Epochly automatically chooses Level 4 for large arrays
large_data = np.random.rand(10_000_000)
result = auto_optimized(large_data)

import epochly
import numpy as np
import time
def benchmark_all_levels(func, *args):
    """Compare all optimization levels"""
    results = []
    
    for level in [0, 1, 2, 3]:
        @epochly.optimize(level=level)
        def optimized_func(*args):
            return func(*args)
        
        # Warmup
        optimized_func(*args)
        
        # Measure
        times = []
        for _ in range(5):
            start = time.perf_counter()
            optimized_func(*args)
            elapsed = time.perf_counter() - start
            times.append(elapsed)
        
        avg_time = np.mean(times)
        results.append({
            'level': level,
            'time': avg_time,
            'speedup': None  # Calculate later
        })
    
    # Calculate speedups
    baseline = results[0]['time']
    for r in results:
        r['speedup'] = baseline / r['time']
    
    return results
# Define workload
def compute_workload(n):
    arr = np.random.rand(n)
    return np.sum(arr ** 2 + np.sin(arr))
# Benchmark
results = benchmark_all_levels(compute_workload, 5_000_000)
# Print results
print("\nLevel Comparison:")
print(f"{'Level':<8} {'Time (s)':<12} {'Speedup':<10}")
print("-" * 35)
for r in results:
    print(f"{r['level']:<8} {r['time']:<12.4f} {r['speedup']:<10.2f}x")

import epochly
import inspect
import numpy as np
def analyze_workload(func, sample_data):
    """Analyze workload and recommend best level"""
    
    # Check function characteristics
    source = inspect.getsource(func)
    
    recommendations = []
    
    # Check for I/O operations
    io_keywords = ['open(', 'read(', 'write(', 'requests.', 'urllib']
    if any(keyword in source for keyword in io_keywords):
        recommendations.append({
            'level': 1,
            'reason': 'I/O operations detected',
            'confidence': 'high'
        })
    
    # Check for loops
    loop_keywords = ['for ', 'while ']
    if any(keyword in source for keyword in loop_keywords):
        recommendations.append({
            'level': 2,
            'reason': 'Loops detected (JIT will help)',
            'confidence': 'medium'
        })
    
    # Check for list comprehensions
    if '[' in source and 'for' in source:
        recommendations.append({
            'level': 3,
            'reason': 'List comprehension (parallelizable)',
            'confidence': 'medium'
        })
    
    # Check data size
    if hasattr(sample_data, '__len__'):
        size = len(sample_data)
        if size > 1_000_000:
            recommendations.append({
                'level': 4,
                'reason': f'Large data size ({size:,} elements)',
                'confidence': 'high' if size > 10_000_000 else 'medium'
            })
    
    # Benchmark to verify
    best_level = 0
    best_time = float('inf')
    
    for rec in recommendations:
        level = rec['level']
        
        with epochly.optimize_context(level=level):
            # Warmup
            func(sample_data)
            
            # Measure
            import time
            start = time.perf_counter()
            func(sample_data)
            elapsed = time.perf_counter() - start
            
            if elapsed < best_time:
                best_time = elapsed
                best_level = level
    
    return {
        'recommended_level': best_level,
        'recommendations': recommendations,
        'benchmark_time': best_time
    }
# Example usage
def my_workload(data):
    result = 0
    for x in data:
        result += x ** 2
    return result
sample = np.random.rand(100_000)
analysis = analyze_workload(my_workload, sample)
print(f"Recommended Level: {analysis['recommended_level']}")
print("\nReasons:")
for rec in analysis['recommendations']:
    print(f"  Level {rec['level']}: {rec['reason']} (confidence: {rec['confidence']})")

# Start with automatic selection
@epochly.optimize()
def my_function(data):
    return process(data)
# Or use Level 2 as a safe default for CPU-bound work
@epochly.optimize(level=2)
def cpu_function(data):
    return compute(data)

import epochly
# Test all applicable levels
levels_to_test = [0, 1, 2, 3]
for level in levels_to_test:
    with epochly.benchmark_context(f"Level {level}"):
        with epochly.optimize_context(level=level):
            result = my_function(data)
# Choose the fastest

# Small data: Overhead may outweigh benefits
small_data = np.random.rand(100)
# Use Level 0 or no optimization for small data
@epochly.optimize(level=0)
def process_small(data):
    return np.sum(data)
# Large data: Optimization pays off
large_data = np.random.rand(10_000_000)
# Use Level 3 for large parallel workloads
@epochly.optimize(level=3)
def process_large(data):
    return [np.sum(chunk) for chunk in chunks(data)]

import epochly
# Use Level 0 for monitoring without optimization
@epochly.optimize(level=0)
def production_function(data):
    result = process(data)
    
    # Log metrics
    metrics = epochly.get_metrics()
    log_to_monitoring(metrics)
    
    return result

import epochly
import numpy as np
# ❌ MISTAKE: Level 3 for small data
@epochly.optimize(level=3)
def bad_small_data():
    data = [i ** 2 for i in range(100)]  # Only 100 items!
    return sum(data)
# ✅ CORRECT: No optimization for small data
def good_small_data():
    data = [i ** 2 for i in range(100)]
    return sum(data)
# ❌ MISTAKE: Level 2 for I/O
@epochly.optimize(level=2)
def bad_io_operation():
    return [open(f).read() for f in file_paths]
# ✅ CORRECT: Level 1 for I/O
@epochly.optimize(level=1)
def good_io_operation():
    return [open(f).read() for f in file_paths]
# ❌ MISTAKE: Level 4 without checking GPU
@epochly.optimize(level=4)
def bad_gpu_usage():
    return np.sum(small_array)
# ✅ CORRECT: Check GPU and data size
@epochly.optimize(level=4)
def good_gpu_usage():
    # Only use if GPU available and data is large
    if epochly.is_gpu_available() and len(data) > 1_000_000:
        return np.sum(data)
    else:
        raise ValueError("GPU not available or data too small")

┌─────────────────────────────────────┐
│ What type of workload?              │
└────────────┬────────────────────────┘
             │
    ┌────────┴────────┐
    │                 │
    ▼                 ▼
┌─────────┐      ┌─────────┐
│ I/O     │      │ CPU     │
│ bound?  │      │ bound?  │
└────┬────┘      └────┬────┘
     │                │
     ▼                ▼
 Level 1         ┌─────────┐
 Threading       │ Loops?  │
                 └────┬────┘
                      │
         ┌────────────┴────────────┐
         │                         │
         ▼                         ▼
    ┌─────────┐             ┌──────────┐
    │ Yes     │             │ No       │
    └────┬────┘             └────┬─────┘
         │                       │
         ▼                       ▼
   ┌──────────┐           ┌──────────┐
   │ Level 2  │           │ Parallel?│
   │ JIT      │           └────┬─────┘
   └──────────┘                │
                    ┌───────────┴────────┐
                    │                    │
                    ▼                    ▼
              ┌──────────┐         ┌─────────┐
              │ Yes      │         │ Large   │
              └────┬─────┘         │ arrays? │
                   │               └────┬────┘
                   ▼                    │
             ┌──────────┐               ▼
             │ Level 3  │         ┌──────────┐
             │ Multicore│         │ Level 4  │
             └──────────┘         │ GPU      │
                                  └──────────┘