基于权重的变量选择与按权重比例随机分配实现方案（网站并行营销活动场景）

Great question! For your use case—needing exact selection counts per variable over 100 runs while keeping the order random—here's a straightforward, reliable approach tailored to your marketing/A/B testing scenario:

Core Approach #

The key is to pre-build a "weighted pool" of variables where each entry appears exactly as many times as its weight, then shuffle this pool to randomize the selection order. This guarantees exact counts and ensures no predictable sequence.

Step-by-Step Implementation #

1. Build the Weighted Pool
   Create a list where each variable is repeated exactly according to its weight. For your example:

   • v1 → 30 entries
   • v2 → 20 entries
   • v3 → 10 entries
   • v4 → 25 entries
   • v5 → 15 entries
     The total length of this list will be 100, matching your 100 runs.

2. Shuffle the Pool
   Randomly rearrange the list to eliminate any fixed order. This ensures each variable's selections are spread randomly across the 100 runs.

3. Select Sequentially
   For each run, pick the next element from the shuffled list. Since the pool has exactly the required number of each variable, you'll hit your exact count targets every time.

Example Code (Python) #

Here's a practical implementation you can adapt to your programming language:

import random
from collections import Counter

# Define your variables and weights
variable_weights = {
    "v1": 30,
    "v2": 20,
    "v3": 10,
    "v4": 25,
    "v5": 15
}

# Create the weighted pool
weighted_pool = []
for var, count in variable_weights.items():
    weighted_pool.extend([var] * count)

# Shuffle to randomize order
random.shuffle(weighted_pool)

# Simulate 100 selections (just iterate through the shuffled list)
selections = weighted_pool.copy()

# Verify the counts (optional, for testing)
print("Selection counts:", Counter(selections))

Why This Works #

• Exact Counts: By explicitly including each variable the required number of times, you eliminate any probabilistic variation that might occur with random number range checks (which could give 29 or 31 for v1 instead of exactly 30).
• Random Order: Shuffling the pool ensures that the selection sequence is unpredictable—each permutation of the list is equally likely, so there's no bias in when variables are chosen.
• Scalability: For larger weights (e.g., summing to 10,000 sessions), this method is still efficient unless memory is a strict constraint (in which case you can use a weighted shuffle algorithm that doesn't require building the full list).

Application to Your Marketing Use Case #

This approach is ideal for your parallel marketing campaigns or A/B tests. For example, if you have two campaigns with 50/50 weights:

• Build a list with each campaign 50 times
• Shuffle it once
• Assign each incoming session to the next element in the list
  This guarantees that exactly 50% of sessions see each campaign, even with smaller sample sizes where probabilistic methods might drift from the target.

内容的提问来源于stack exchange，提问作者Mohsin Sethi