speed_test.py

#!/usr/bin/env python3
"""
Speed testing script for NFL play detection system.
Tests both individual clip processing and batch processing performance.
"""
import time
import os
import glob
from inference import predict_clip, transcribe_clip, analyze_play_state
from config import DEFAULT_DATA_DIR

def test_single_clip_speed(clip_path: str, num_runs: int = 5):
    """Test speed of processing a single clip multiple times"""
    print(f"\n=== Single Clip Speed Test ===")
    print(f"Testing: {os.path.basename(clip_path)}")
    print(f"Runs: {num_runs}")
    
    video_times = []
    audio_times = []
    
    for i in range(num_runs):
        print(f"Run {i+1}/{num_runs}...")
        
        # Test video classification speed
        start_time = time.time()
        video_results = predict_clip(clip_path)
        video_time = time.time() - start_time
        video_times.append(video_time)
        
        # Test audio transcription speed  
        start_time = time.time()
        audio_result = transcribe_clip(clip_path)
        audio_time = time.time() - start_time
        audio_times.append(audio_time)
        
        # Test play state analysis speed
        start_time = time.time()
        play_state, confidence = analyze_play_state(video_results)
        analysis_time = time.time() - start_time
        
        total_time = video_time + audio_time + analysis_time
        print(f"  Video: {video_time:.2f}s | Audio: {audio_time:.2f}s | Analysis: {analysis_time:.3f}s | Total: {total_time:.2f}s")
    
    # Calculate averages
    avg_video = sum(video_times) / len(video_times)
    avg_audio = sum(audio_times) / len(audio_times)
    avg_total = avg_video + avg_audio
    
    print(f"\n=== Average Times ===")
    print(f"Video Classification: {avg_video:.2f}s")
    print(f"Audio Transcription: {avg_audio:.2f}s") 
    print(f"Total per clip: {avg_total:.2f}s")
    print(f"Estimated clips/minute: {60/avg_total:.1f}")
    
    return avg_video, avg_audio, avg_total

def test_batch_speed(input_dir: str = DEFAULT_DATA_DIR, max_clips: int = 10):
    """Test speed of batch processing"""
    print(f"\n=== Batch Processing Speed Test ===")
    
    # Find clips
    patterns = ["*.mov", "*.mp4"]
    clips = []
    for pat in patterns:
        clips.extend(glob.glob(os.path.join(input_dir, pat)))
    clips = sorted(clips)[:max_clips]  # Limit for speed testing
    
    if not clips:
        print(f"No clips found in '{input_dir}'")
        return
    
    print(f"Processing {len(clips)} clips...")
    
    total_start = time.time()
    clip_times = []
    
    for i, clip in enumerate(clips):
        clip_name = os.path.basename(clip)
        print(f"[{i+1}/{len(clips)}] {clip_name}")
        
        clip_start = time.time()
        
        # Process clip (same as run_all_clips.py but with timing)
        video_results = predict_clip(clip)
        play_state, confidence = analyze_play_state(video_results)
        transcript = transcribe_clip(clip)
        
        clip_time = time.time() - clip_start
        clip_times.append(clip_time)
        
        print(f"  Time: {clip_time:.2f}s | Play State: {play_state} ({confidence:.3f})")
    
    total_time = time.time() - total_start
    avg_clip_time = sum(clip_times) / len(clip_times) if clip_times else 0
    
    print(f"\n=== Batch Results ===")
    print(f"Total clips: {len(clips)}")
    print(f"Total time: {total_time:.2f}s")
    print(f"Average per clip: {avg_clip_time:.2f}s")
    print(f"Clips per minute: {60/avg_clip_time:.1f}")
    print(f"Time for 100 clips: ~{(avg_clip_time * 100)/60:.1f} minutes")
    
    return total_time, avg_clip_time

def test_pipeline_modes(input_dir: str = DEFAULT_DATA_DIR, max_clips: int = 5):
    """Compare different pipeline processing modes"""
    print(f"\n=== Pipeline Mode Comparison ===")
    
    # Find clips
    patterns = ["*.mov", "*.mp4"]
    clips = []
    for pat in patterns:
        clips.extend(glob.glob(os.path.join(input_dir, pat)))
    clips = sorted(clips)[:max_clips]
    
    if not clips:
        print(f"No clips found in '{input_dir}'")
        return
    
    # Test 1: Video-only processing
    print(f"\n🚀 Video-Only Processing ({len(clips)} clips)")
    print("-" * 50)
    video_start = time.time()
    video_times = []
    
    for i, clip in enumerate(clips, 1):
        clip_name = os.path.basename(clip)
        clip_start = time.time()
        
        scores = predict_clip(clip)
        play_state, confidence = analyze_play_state(scores)
        
        clip_time = time.time() - clip_start
        video_times.append(clip_time)
        print(f"[{i}/{len(clips)}] {clip_name}: {clip_time:.2f}s | {play_state}")
    
    video_total = time.time() - video_start
    video_avg = video_total / len(clips)
    
    # Test 2: Audio-only processing  
    print(f"\n🎙️  Audio-Only Processing ({len(clips)} clips)")
    print("-" * 50)
    audio_start = time.time()
    audio_times = []
    
    for i, clip in enumerate(clips, 1):
        clip_name = os.path.basename(clip)
        clip_start = time.time()
        
        transcript = transcribe_clip(clip)
        
        clip_time = time.time() - clip_start
        audio_times.append(clip_time)
        print(f"[{i}/{len(clips)}] {clip_name}: {clip_time:.2f}s")
    
    audio_total = time.time() - audio_start
    audio_avg = audio_total / len(clips)
    
    # Results comparison
    combined_avg = video_avg + audio_avg
    
    print(f"\n📊 Pipeline Comparison Results")
    print("=" * 50)
    print(f"Video-only processing:")
    print(f"  Average per clip: {video_avg:.2f}s")
    print(f"  Throughput: {60/video_avg:.1f} clips/minute")
    print(f"  100 clips: ~{video_avg*100/60:.1f} minutes")
    
    print(f"\nAudio-only processing:")
    print(f"  Average per clip: {audio_avg:.2f}s")  
    print(f"  Throughput: {60/audio_avg:.1f} clips/minute")
    print(f"  100 clips: ~{audio_avg*100/60:.1f} minutes")
    
    print(f"\nCombined (sequential phases):")
    print(f"  Total per clip: {combined_avg:.2f}s")
    print(f"  Throughput: {60/combined_avg:.1f} clips/minute")
    print(f"  100 clips: ~{combined_avg*100/60:.1f} minutes")
    
    print(f"\nSpeed advantage of video-only: {audio_avg/video_avg:.1f}x faster")
    
    return {
        "video_avg": video_avg,
        "audio_avg": audio_avg,
        "combined_avg": combined_avg,
        "speedup": audio_avg/video_avg
    }

def main():
    print("🏈 NFL Play Detection Speed Test")
    print("=" * 50)
    
    # Find a test clip
    test_clips = glob.glob("data/*.mov") + glob.glob("data/*.mp4")
    if not test_clips:
        print("❌ No clips found in data/ directory")
        return
    
    test_clip = test_clips[0]
    
    # Single clip speed test
    avg_video, avg_audio, avg_total = test_single_clip_speed(test_clip, num_runs=3)
    
    # Batch speed test
    total_time, avg_clip_time = test_batch_speed(max_clips=5)
    
    # Pipeline comparison test
    pipeline_results = test_pipeline_modes(max_clips=5)
    
    print(f"\n🎯 Performance Summary")
    print(f"=" * 30)
    print(f"Single clip average: {avg_total:.2f}s")
    print(f"Batch average: {avg_clip_time:.2f}s")
    print(f"Estimated throughput: {60/avg_clip_time:.1f} clips/minute")
    
    # Estimate processing time for different scenarios
    print(f"\n📊 Processing Time Estimates:")
    print(f"  10 clips: ~{(avg_clip_time * 10)/60:.1f} minutes")
    print(f"  50 clips: ~{(avg_clip_time * 50)/60:.1f} minutes") 
    print(f"  100 clips: ~{(avg_clip_time * 100)/60:.1f} minutes")
    print(f"  1 hour of 2s clips (1800): ~{(avg_clip_time * 1800)/60:.0f} minutes")
    
    print(f"\n🚀 Optimization Recommendations:")
    print("=" * 40)
    print(f"For continuous pipelines:")
    print(f"  • Use --video-only for {pipeline_results['speedup']:.1f}x faster processing")
    print(f"  • Process video in real-time, audio in batch later")
    print(f"  • Video-only: {60/pipeline_results['video_avg']:.1f} clips/min vs {60/pipeline_results['combined_avg']:.1f} clips/min full pipeline")

if __name__ == "__main__":
    main()