R语言正态似然函数最大化代码中params[!fixed] <- p行含义解析

Hey there! Let's break down that line of code and its role in maximizing the normal likelihood function step by step.

1. 先搞懂params[!fixed] <- p到底在做什么 #

First, let's contextualize where this line lives: it's inside the inner function returned by make.NegLogLik, which is the actual negative log-likelihood function we pass to optimizers like optim or optimize.

Let's unpack each part:

• The fixed argument in the outer make.NegLogLik function lets you specify which parameters (μ and σ, corresponding to positions 1 and 2 in the vector) are fixed to a constant value and which are free to be optimized:
  • If an element is FALSE, that parameter is a free variable we want the optimizer to find
  • If an element is a number (like 2 in your second example), that parameter is locked to that value and won't change during optimization
• params <- fixed initializes our full parameter vector with the fixed values you specified; free parameter positions start as FALSE temporarily
• !fixed converts the fixed vector into a logical mask: any FALSE (free parameter) becomes TRUE, and any numeric value (fixed parameter) becomes FALSE (since non-zero numbers count as TRUE in R, so !2 is FALSE)
• Finally, params[!fixed] <- p takes the candidate parameter values p passed in by the optimizer and plugs them into the free parameter positions in params. Now params is a complete vector with both fixed values and the current candidate values for free parameters, ready for the likelihood calculation.

2. 这行代码在最大化正态似然函数中的核心作用 #

记住：最小化负对数似然（这是optim这类优化器的默认行为）和最大化对数似然（进而最大化似然函数本身）是完全等价的。这行代码解决了一个关键问题：它让我们不管是优化一个参数、两个参数，甚至固定所有参数，都能复用同一套核心似然计算代码。

结合你的例子具体看：

例子1：同时优化μ和σ #

nLL <- make.NegLogLik(normals)
optim(c(mu = 0, sigma = 1), nLL)$par

这里fixed = c(FALSE, FALSE)，所以!fixed会变成c(TRUE, TRUE)。优化器传入一个2元素的向量p（候选的μ和σ值），params[!fixed] <- p会把这两个值分别填入params的对应位置。函数随后计算该(μ, σ)组合对应的负对数似然，优化器则在二维参数空间中搜索能最小化这个值的参数组合。

例子2：固定σ，只优化μ #

nLL <- make.NegLogLik(normals, c(FALSE, 2))
optimize(nLL, c(-1, 3))$minimum

这里fixed = c(FALSE, 2)，所以!fixed是c(TRUE, FALSE)。此时优化器只传入单个值p（候选的μ），params[!fixed] <- p会把这个值赋值给params的第一个位置，而第二个位置保持固定的2。因为只需要在一维空间搜索，我们可以用更简单的optimize函数，且核心的似然计算代码完全不需要修改。

简单来说，这行代码就像一个灵活的"适配器"：一边对接只关心自由参数的优化器，另一边对接需要完整参数集（固定+自由）的似然计算逻辑。

内容的提问来源于stack exchange，提问作者Shreyas