Python新手求助：基于偏移钟形曲线生成random.choices权重的实现方案

Hey there! Let's break down how to solve this problem—you're onto a solid idea with using a skewed bell curve to weight your choices, and we can make this work with or without external libraries, depending on what you prefer.

First, let's align on the basics: you have 5 options (a to e) mapped to x-values [0.2, 0.4, 0.6, 0.8, 1.0], and you want a skewed bell curve where the peak sits exactly at your parameter p (ranging from 0.1 to 1). We'll generate weights for each option using this curve, then feed them to random.choices to build your list q.

Option 1: Using scipy.stats.skewnorm (More Precise Skew) #

If you don't mind using the scipy library, this gives you a proper skewed normal distribution tailored to your needs. Here's how to tweak it so the peak lands exactly at p:

import random
import numpy as np
from scipy.stats import skewnorm

def generate_skewnorm_weights(p, x_points=[0.2, 0.4, 0.6, 0.8, 1.0], scale=0.2):
    # Calculate skew direction/strength: negative for left-skew (p < 0.5), positive for right-skew (p > 0.5)
    # Adjust the multiplier (10) to make skew stronger or milder
    skew = (p - 0.5) * 10
    
    # Skewnorm's peak isn't at its mean, so we calculate the correct mean (loc) to center the peak at p
    loc = p + (skew * scale) / np.sqrt(1 + skew**2)
    
    # Create the skewed normal distribution
    dist = skewnorm(a=skew, loc=loc, scale=scale)
    
    # Get probability density values for each x-point
    pdf_values = dist.pdf(x_points)
    
    # Normalize to get weights that sum to 1 (easier to reason about probabilities)
    weights = pdf_values / pdf_values.sum()
    return weights

# Example usage
w = ['a', 'b', 'c', 'd', 'e']
p = 0.55  # Peak should be at 'c'
weights = generate_skewnorm_weights(p)
q = random.choices(w, weights=weights, k=10)
print(q)

Quick Adjustments: #

• scale: Smaller values make the peak sharper (weights cluster tightly around the top option), larger values make weights more evenly distributed.
• Skew multiplier: Change 10 to 5 for milder skew, or 15 for more extreme, lopsided distributions.

Option 2: Custom Skewed Bell Curve (No scipy Needed) #

If you want to keep things lightweight and avoid external libraries, we can build a simple skewed bell curve using basic numpy functions. This is perfect for beginners who want to stick to standard libraries:

import random
import numpy as np

def custom_skewed_bell(x, p, scale=0.2, skew_strength=2):
    # Base Gaussian curve for the core bell shape
    gaussian = np.exp(-((x - p)/scale)**2)
    # Skew term: amplifies values on the side of p, weakens the opposite side
    skew_term = np.where(x > p, 1 + skew_strength*(x - p), 1 - skew_strength*(p - x))
    # Ensure we don't get negative values (just a safe guard)
    return np.maximum(gaussian * skew_term, 0)

def generate_custom_weights(p, x_points=[0.2, 0.4, 0.6, 0.8, 1.0], scale=0.2, skew_strength=2):
    bell_values = custom_skewed_bell(np.array(x_points), p, scale, skew_strength)
    # Normalize weights to sum to 1
    weights = bell_values / bell_values.sum()
    return weights

# Example usage
w = ['a', 'b', 'c', 'd', 'e']
p = 0.1  # Peak should be at 'a'
weights = generate_custom_weights(p)
q = random.choices(w, weights=weights, k=10)
print(q)

Quick Adjustments: #

• scale: Same as above—controls how narrow or wide the bell curve is.
• skew_strength: Higher values make the skew more dramatic (e.g., 3 will push most weight to the peak side), while 1 makes skew very subtle.

Key Notes #

• Normalization: We always normalize weights to sum to 1. This isn't strictly required for random.choices (it accepts any positive weights), but it makes it much easier to understand the approximate probability of each option.
• p Alignment: The x-values [0.2, 0.4, ..., 1.0] perfectly match your p range (0.1 to 1). For example, p=0.55 lands closest to 0.6 (option c), which exactly matches your requirement of c being the most probable.

内容的提问来源于stack exchange，提问作者Kit McCarthy