Non-Parametric Model Evaluation

Module by: Don Johnson

Note: Your browser may not currently support MathML. Chrome will attempt to display MathML, but it should not be considered correct. Safari will attempt to display MathML, but it should not be considered correct. Please install the required MathPlayer plugin to view MathML correctly.Firefox requires additional mathematics fonts to display MathML correctly.Firefox requires additional mathematics fonts to display MathML correctly. See our browser support page for additional details. You can always view the correct math in the PDF version.

(Show MathML examples)

If the two examples below are mathematically the same, your browser is correctly displaying MathML, and you can dismiss this message.

Example 1:	$\frac{\sqrt{\frac{1}{2}}}{\sqrt{{f(x+y)}^2}}\frac{\sqrt{1}}{2}\langle ℝ\rangle$	Example 2:		(Hide examples)

If very uncertain about model accuracy, assuming a form for the nominal density may be questionable or quantifying the degree of uncertainty may be unreasonable. In these cases, any formula for the underlying probability densities may be unjustified, but the model evaluation problem remains. For example, we may want to determine a signal presence or absence in an array output (non-zero mean vs. zero mean) without much knowledge of the contaminating noise. If minimal assumptions can be made about the probability densities, non-parametric model evaluation can be used (Gibson and Melsa). In this theoretical framework, no formula for the conditional densities is required; instead, we use worst-case densities which conform to the weak problem specification. Because few assumptions about the probability models are used, non-parametric decision rules are robust: they are insensitive to modeling assumptions because so few are used. The "robust" test of the previous section are so-named because they explicitly encapsulate model imprecision. In either case, one should expect greater performance (smaller error probabilities) in non-parametric decision rules than possible from a "robust" one.

Two hypothesized models are to be tested; ℳ 0 ℳ 0 is intended to describe the situation "the observed data have zero mean" and the other "a non-zero mean is present." We make the usual assumption that the LL observed data values are statistically independent. The only assumption we will make about the probabilistic descriptions underlying these models is that the median of the observations is zero in the first instance and non-zero in the second. The median of a random variable is the "half-way" value: the probability that the random variable is less than the median is one-half as is the probability that it is greater. The median and mean of a random variable are not necessarily equal; for the special case of a symmetric probability density they are. In any case, the non-parametric models will be stated in terms of the probability that an observation is greater than zero. ℳ 0 : Pr r l ≥0=12 ℳ 0 : r l 0 1 2 ℳ 1 : Pr r l ≥0>12 ℳ 1 : r l 0 1 2 The first model is equivalent to a zero-median model for the data; the second implies that the median is greater than zero. Note that the form of the two underlying probability densities need not be the same to correspond to the two models; they can differ in more general ways than in their means.

To solve this model evaluation problem, we seek (as do all robust techniques) the worst-case density, the density satisfying the conditions for one model that is maximally difficult to distinguish from a given density under the other. Several interesting problems arise in this approach. First of all, we seek a non-parametric answer: the solution must not depend on unstated parameters (we should not have to specify how large the non-zero mean might be). Secondly, the model evaluation rule must not depend on the form for the given density. These seemingly impossible properties are easily satisfied. To find the worst-case density, first define p + r l | ℳ 1 r l p + r l ℳ 1 r l to be the probability density of the l th l th observation assuming that ℳ 1 ℳ 1 is true and that the observation was non-negative. A similar definition for negative values is needed. p + r l | ℳ 1 r l =p r l | ℳ 1 r l ≥0 r l p + r l ℳ 1 r l p r l ℳ 1 r l 0 r l p - r l | ℳ 1 r l =p r l | ℳ 1 r l <0 r l p - r l ℳ 1 r l p r l ℳ 1 r l 0 r l In terms of these quantities, the conditional density of an observation under ℳ 1 ℳ 1 is given by p r l | ℳ 1 r l =Pr r l ≥0| ℳ 1 p + r l | ℳ 1 r l +1−Pr r l ≥0| ℳ 1 p - r l | ℳ 1 r l p r l ℳ 1 r l ℳ 1 r l 0 p + r l ℳ 1 r l 1 ℳ 1 r l 0 p - r l ℳ 1 r l The worst-case density under ℳ 0 ℳ 0 would have exactly the same functional form as this one for positive and negative values while having a zero median. 1 As depicted in Figure 1, a density meeting these requirements is p r l | ℳ 1 r l = p + r l | ℳ 1 r l + p - r l | ℳ 1 r l 2 p r l ℳ 1 r l p + r l ℳ 1 r l p - r l ℳ 1 r l 2

Figure 1: For each density having a positive-valued median, the worst-case density having zero median would have exactly the same functional form as the given one on the positive and negative real lines, but with the areas adjusted to be equal. Here, a unit-mean, unit-variance Gaussian density and its corresponding worst-case density is usually discontinuous at the origin; be that as it may, this rather bizarre worst-case density leads to a simple non-parametric decision rule.

The likelihood ratio for a single observation would be 2Pr r l ≥0| ℳ 1 2 ℳ 1 r l 0 for non-negative values and 21−Pr r l ≥0| ℳ 1 2 1 ℳ 1 r l 0 for negative values. While the likelihood ratio depends on Pr r l ≥0| ℳ 1 ℳ 1 r l 0 , which is not specified in out non-parametric model, the sufficient statistic will not depend on it! To see this, note that the likelihood ratio varies only with the sign of the observation. Hence, the optimal decision rule amounts to counting how many of the observations are positive; this count can be succinctly expressed with the unit-step function u· u · as ∑l=0L−1u r l l 0 L 1 u r l . 2 Thus, the likelihood ratio for the LL statistically independent observations is written 2LPr r l ≥0| ℳ 1 ∑lu r l 1−Pr r l ≥0| ℳ 1 L−∑lu r l ≷ ℳ 0 ℳ 1 η 2 L ℳ 1 r l 0 l u r l 1 ℳ 1 r l 0 L l u r l ≷ ℳ 0 ℳ 1 η Making the usual simplifications, the unknown probability Pr r l ≥0| ℳ 1 ℳ 1 r l 0 can be maneuvered to the right side and merged with the threshold. The optimal non-parametric decision rule thus compares the sufficient statistic - the count of positive-valued observations - with a threshold determined by the Neyman-Pearson criterion.

To find the threshold γγ, we can use the Central Limit Theorem to approximate the probability distribution of the sum by a Gaussian. Under ℳ 0 ℳ 0 , the expected value of u r l u r l is 12 1 2 and the variance is 14 1 4 . To the degree that the Central Limit Theorem reflects the false-alarm probability (see this problem), P F P F is approximately given by P F =Qγ−L2L4 P F Q γ L 2 L 4 and the threshold is found to be γ=L2Q-1 P F +L2 γ L 2 Q P F L 2 As it makes no sense for the threshold to be greater than LL (how many positively values observations can there be?), the specified false-alarm probability must satisfy P F ≥QL P F Q L . This restriction means that increasing stringent requirements on the false-alarm probability can only be met if we have sufficient data.

Comments, questions, feedback, criticisms?

Send feedback

• E-mail the author of the module, Non-Parametric Model Evaluation

Footer

This work is licensed by Don Johnson under a Creative Commons Attribution License (CC-BY 1.0), and is an Open Educational Resource.

Last edited by Charlet Reedstrom on Sep 15, 2003 1:41 pm GMT-5.