DOCSLIB.ORG

A Loss Function Analysis for Classification Methods in Text Categorization

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Loss Function Analysis for Classification Methods in Text Categorization 472 A Loss Function Analysis for Classification Methods in Text Categorization Fan LI HUSTLF~CS.CMU.EDU Carnegie Mellon Univ, 4502 NSH, 5000 Forbes Avenue, Pittsburgh, PA 15213 USA Yiming Yang [email protected] Carnegie Mellon Univ, 4502 NSH, 5000 Forbes Avenue, Pittsburgh, PA 15213 USA Abstract Support Vector Machine (SVM)and other approaches (Yang & Liu, 1999; Yang, 1999; Joachims, 1998; Mc- This paper presents a formal analysis of pop- Callum & Nigam; Zhang & Oles, 2001; Lewis et al., ular text classification methods, focusing on 2003). As more methods are published, we need to their loss functions whoseminimization is es- have a sound theoretical framework for cross-method sential to the optimization of those methods, comparison. Recent work in machine learning focus- and whose decomposition into the training- ing on the regularization of classification methodsand set loss and the model complexity enables on the analysis of their loss functions is a step in this cross-method comparisons on a commonba- direction. sis from an optimization point of view. Those methods include Support Vector Machines, Vapnik (Vapnik, 1995) defined the objective function Linear Regression, Logistic Regression, Neu- in SVMas minimizing the expected risk on test exam- ral Network, Naive Baycs, K-Nearest Neigh- ples, and decomposedthat risk into two components: bor, Rocchio-style and Multi-class Prototype the empirical risk that reflects the training-set errors classifiers. Theoretical analysis (including of the classifier, and the inverse of marginwidth that our new derivations) is provided for each reflects howfar the positive and negative training ex- method,along with e~-aluation results for all amples of a category are separated by the decision sur- the methods on the Reuters-21578 bench- face. Thus, both the minimization of training-set er- mark corpus. Using linear regression, neural rors and the maximization of the margin width are the networks and logistic regression methods as criteria used in the optimization of SVM.Balancing examples, we show that properly tuning the betweenthe two criteria has been referred as the regu- balance betweenthe training-set loss and the larization of a classifier; the degree of regularization is complexity penalty would have a significant often controlled by a parameter in that method (sec- impact to the performanceof a classifier. In tion 2). SVMhave been extremely successful in text linear regression, in particular, the tuning of categorization, often resulting in the best performance the complexity penalty yielded a result (mea- in benchmark evaluations (Joachims, 1998; Yang sured using macro-averaged F1) that outper- Liu, 1999; Lewiset al., 2003). formed all text categorization methods ever Hastie et al. (Hastie et al. 2001) presented a moregen- evaluated on that benchmarkcorpus, includ- eral frameworkfor estimating the potential of a model ing Support Vector Machines. in makingclassification errors, and used a slightly dif- ferent terminology: loss or generalization error corre- spondingto the expected risk, training-set loss corre- 1. Introduction sponding to the empirical risk, and model complexity Text categorization is an active research area in ma~ corresponding to the margin-related risk in SVM.Us- ing this framework they comparedalternative ways to chine learning and information retrieval. A large num- ber of statistical classification methodshave been ap- penalize the model complexity, including the Akalke plied to this problem, including linear regression, lo- Information Criterion (AIC), the Bayesian Informa- gistic regression (LR), neural networks (NNet), Naive tion Criterion (BIC), and the MinimumDescription Bayes (NB), k-nearest neighbor (kNN), Rocchio-style, Length (MDL)criterion. More interestingly, they Proceedingsof the Twentieth International Conferenceon MachineLearning (ICML-2003),Washington DC, 2003. 473 comparedthe differences in the training-set loss func- represents the values of the p input variables in the tions for SVM,LLSF, LR and AdaBoost, in a way such ith training example. Scalar Yl E {-1,1} (unless that the sensitivity of those methods with respect to otherwise specified) is the class label. classification errors on training examplescan be easily compared(section 2). ¯ Vector ~ = (ill,...,/~v) T consists of the parame- ters in a linear classifier, whichare estimated us- It wouldbe valuable to analyze a broader range of clas- ing the training data. sification methodsin a similar fashion as presented by Hestie et al., so that the comparison amongmethods ¯ Alinea scalar f(a~i,/~) = ~’i/~is the classifier’s out- can be madeexplicitly in terms of their inductive bi- put given input gi, and the quantity Yif(~i,/~) ases with respect to training examples, or in terms of shows how much the system’s output agrees with their penalty functions for model complexity. For this the truth label we need a formal analysis on the optimization criterion of each method, in the form of a loss function that ¯ The 2-norm of/~ is represented as [[/~1[ and the decomposes into the training-set error term and the 1-normof/~ is represented as IIEII model complexity term. Such a formal analysis, how- ever, often is not available in the literature for popular Note that we purposely chose to define gi as a hori- text categorization methods, such as Nave Bayes, kNN zontal vector and fl as a vertical vector, so that we can and Rocchio-style classifiers. conveniently write xi~ for the dot product ~-]~=1x~/~k The primary contribution we offer here is a loss- (and vice versa), which will be frequently seen in our function based study for eight classifiers popular in derivations. text categorization, including SVM,linear regression, logistic regression, neural networks, Rocchio-style, 2.1. SVM Prototypes, kNN and Nave Bayes. We provide our SVMhas been extremely successful in text categoriza- own derivations for the loss function decomposition tion. Multiple versions of SVMexist; in this paper we in Rocchio-style, NB, kNNand multi-class prototypes only use linear SVMfor our analysis, partly for clarity (Prototypes), which have not been reported before. and simplicity of our analysis, and partly because lin- Wealso show the importance of properly tuning the ear SVMperformed as well as other versions of SVMin amount of regularization by using controlled examina- text categorization evaluations(Joachims, 1998). SVM tions of LLSF, LR and NNet with and without reg- emphasizesthe generalization capability of the classi- ularization. Finally, we compare the performance of the eight classifiers with properly tuned regulariza- fier (Hastie et al. 2001), whoseloss function (for class c) has the form of tion (though validation) using a benchmark corpus (Reuters-21578) in text categorization. fl The organization of the remaining parts of this pa- Lc : E(1 - yiEi/~)+ + ~11/~[12 (1) per is aa follows: Section 2 outlines the classifiers and i=1 provides a formal analysis on their loss functions. Sec- in which the training-set loss on a single training ex- tion 3 describes the experiment settings and results. ample is defined to be Section 4 summarizes and offers the concluding re- marks. le = (1-Yi ~ i~)+~ ( 1 =Yi 0 ~i~otherwise wh en Yl Xi~ <_ 1 2. Loss functions of the classifiers The first term in the right hand side of formula 1 is the In order to comparedifferent classifiers on a common cumulative training-set loss and the second term is the basis, we need to present their loss functions in the complexity penalty and both are functions of vector unified form: Lc = gl(YJ(,~i,~)) g2(/~). We cal /L The optimization in SVMis to find/~ that mini- the first term gl (ylf ( Ei, fl) the tr aining-set loss and mizes the sum of the two terms in formula 1. In other the second term g2(/~) the complexity penalty or the words, the optimization in SVMis not only driven by regularizer. The following notation will be used in the the training-set loss, but also driven by the 2-norm rest of this paper: of vector fl, which is determined by the squared sum of the coefficients in /~ and reflects the sensitivity of the mapping function with respect to the input vari- ¯ The training data consL~ts of N pairs of (El,y1), ables. The value of A controls the trade-off between (~’2,y2),...,(X.N,YN). Vector ~i = (Xil,...,Xlp) the two terms, that is, it is the weight (algorithmically 474 determined in the training phase of SVM)of the sec- with the scoring function ](~, fi) = ~fl where fi is the ond term relative to the first term. Formula 1 can be prototype vector, defined to be: transformed into dual form and solved using quadratic 1 b programming. y" = ~-~=~ ~, ~,-~, ~ Ecj ~i ~c This kind of analysis on the loss function in SVMis not new, of course. In fact, it is a part of the SVM : ~1 ~ j, + ~b ~ v,~, (3) theory, and has been presented by other researchers yi:l yi:--I (Vapnik, 1995; Hastie et al. 2001). Our point here where b is a parameter which can be tuned. Nowwe is to start with a good framework and carry out the formal analysis for the other classifiers chosen for this showthat the rcgularized loss function in the Rocchio- style classifier is study in a consistent fashion; someof those classifiers have not been formally analyzed in this manner. Lc =- ~ yi~,ifi bg~ E (4) yi:l y~=--i 2.2. Linear Least Squares Fit (LLSF) In order to minimizethe loss function, we need to take Linear regression, also called Linear Least Squares Fit the first order derivative of formula 4 with respect to (LLSF) in the literature, has performed competitively fi and set it to zero, to SVMand other high performing classifiers (includ- r ing kNNand logistic regression) in text categorization dLc bh evaluations (Yang, 1999). LLSFis similar to linear ,¢ = - ~,yl---1 Y’~’ N~y~=--I~C Y’~’ + Jvd:o SVMin the sense that both leaxn a linear mapping ](~, fi) = ~fi based on tile training data.
Load more

Recommended publications
  • Logistic Regression Trained with Different Loss Functions Discussion
    Logistic Regression Trained with Different Loss Functions Discussion CS6140 1 Notations We restrict our discussions to the binary case. 1 g(z) = 1 + e−z @g(z) g0(z) = = g(z)(1 − g(z)) @z 1 1 h (x) = g(wx) = = w −wx − P wdxd 1 + e 1 + e d P (y = 1jx; w) = hw(x) P (y = 0jx; w) = 1 − hw(x) 2 Maximum Likelihood Estimation 2.1 Goal Maximize likelihood: L(w) = p(yjX; w) m Y = p(yijxi; w) i=1 m Y yi 1−yi = (hw(xi)) (1 − hw(xi)) i=1 1 Or equivalently, maximize the log likelihood: l(w) = log L(w) m X = yi log h(xi) + (1 − yi) log(1 − h(xi)) i=1 2.2 Stochastic Gradient Descent Update Rule @ 1 1 @ j l(w) = (y − (1 − y) ) j g(wxi) @w g(wxi) 1 − g(wxi) @w 1 1 @ = (y − (1 − y) )g(wxi)(1 − g(wxi)) j wxi g(wxi) 1 − g(wxi) @w j = (y(1 − g(wxi)) − (1 − y)g(wxi))xi j = (y − hw(xi))xi j j j w := w + λ(yi − hw(xi)))xi 3 Least Squared Error Estimation 3.1 Goal Minimize sum of squared error: m 1 X L(w) = (y − h (x ))2 2 i w i i=1 3.2 Stochastic Gradient Descent Update Rule @ @h (x ) L(w) = −(y − h (x )) w i @wj i w i @wj j = −(yi − hw(xi))hw(xi)(1 − hw(xi))xi j j j w := w + λ(yi − hw(xi))hw(xi)(1 − hw(xi))xi 4 Comparison 4.1 Update Rule For maximum likelihood logistic regression: j j j w := w + λ(yi − hw(xi)))xi 2 For least squared error logistic regression: j j j w := w + λ(yi − hw(xi))hw(xi)(1 − hw(xi))xi Let f1(h) = (y − h); y 2 f0; 1g; h 2 (0; 1) f2(h) = (y − h)h(1 − h); y 2 f0; 1g; h 2 (0; 1) When y = 1, the plots of f1(h) and f2(h) are shown in figure 1.
    [Show full text]
  • Regularized Regression Under Quadratic Loss, Logistic Loss, Sigmoidal Loss, and Hinge Loss
    Regularized Regression under Quadratic Loss, Logistic Loss, Sigmoidal Loss, and Hinge Loss Here we considerthe problem of learning binary classiers. We assume a set X of possible inputs and we are interested in classifying inputs into one of two classes. For example we might be interesting in predicting whether a given persion is going to vote democratic or republican. We assume a function Φ which assigns a feature vector to each element of x — we assume that for x ∈ X we have d Φ(x) ∈ R . For 1 ≤ i ≤ d we let Φi(x) be the ith coordinate value of Φ(x). For example, for a person x we might have that Φ(x) is a vector specifying income, age, gender, years of education, and other properties. Discrete properties can be represented by binary valued fetures (indicator functions). For example, for each state of the United states we can have a component Φi(x) which is 1 if x lives in that state and 0 otherwise. We assume that we have training data consisting of labeled inputs where, for convenience, we assume that the labels are all either −1 or 1. S = hx1, yyi,..., hxT , yT i xt ∈ X yt ∈ {−1, 1} Our objective is to use the training data to construct a predictor f(x) which predicts y from x. Here we will be interested in predictors of the following form where β ∈ Rd is a parameter vector to be learned from the training data. fβ(x) = sign(β · Φ(x)) (1) We are then interested in learning a parameter vector β from the training data.
    [Show full text]
  • The Central Limit Theorem in Differential Privacy
    Privacy Loss Classes: The Central Limit Theorem in Differential Privacy David M. Sommer Sebastian Meiser Esfandiar Mohammadi ETH Zurich UCL ETH Zurich [email protected] [email protected] [email protected] August 12, 2020 Abstract Quantifying the privacy loss of a privacy-preserving mechanism on potentially sensitive data is a complex and well-researched topic; the de-facto standard for privacy measures are "-differential privacy (DP) and its versatile relaxation (, δ)-approximate differential privacy (ADP). Recently, novel variants of (A)DP focused on giving tighter privacy bounds under continual observation. In this paper we unify many previous works via the privacy loss distribution (PLD) of a mechanism. We show that for non-adaptive mechanisms, the privacy loss under sequential composition undergoes a convolution and will converge to a Gauss distribution (the central limit theorem for DP). We derive several relevant insights: we can now characterize mechanisms by their privacy loss class, i.e., by the Gauss distribution to which their PLD converges, which allows us to give novel ADP bounds for mechanisms based on their privacy loss class; we derive exact analytical guarantees for the approximate randomized response mechanism and an exact analytical and closed formula for the Gauss mechanism, that, given ", calculates δ, s.t., the mechanism is ("; δ)-ADP (not an over- approximating bound). 1 Contents 1 Introduction 4 1.1 Contribution . .4 2 Overview 6 2.1 Worst-case distributions . .6 2.2 The privacy loss distribution . .6 3 Related Work 7 4 Privacy Loss Space 7 4.1 Privacy Loss Variables / Distributions .
    [Show full text]
  • Bayesian Classifiers Under a Mixture Loss Function
    Hunting for Significance: Bayesian Classifiers under a Mixture Loss Function Igar Fuki, Lawrence Brown, Xu Han, Linda Zhao February 13, 2014 Abstract Detecting significance in a high-dimensional sparse data structure has received a large amount of attention in modern statistics. In the current paper, we introduce a compound decision rule to simultaneously classify signals from noise. This procedure is a Bayes rule subject to a mixture loss function. The loss function minimizes the number of false discoveries while controlling the false non discoveries by incorporating the signal strength information. Based on our criterion, strong signals will be penalized more heavily for non discovery than weak signals. In constructing this classification rule, we assume a mixture prior for the parameter which adapts to the unknown spar- sity. This Bayes rule can be viewed as thresholding the \local fdr" (Efron 2007) by adaptive thresholds. Both parametric and nonparametric methods will be discussed. The nonparametric procedure adapts to the unknown data structure well and out- performs the parametric one. Performance of the procedure is illustrated by various simulation studies and a real data application. Keywords: High dimensional sparse inference, Bayes classification rule, Nonparametric esti- mation, False discoveries, False nondiscoveries 1 2 1 Introduction Consider a normal mean model: Zi = βi + i; i = 1; ··· ; p (1) n T where fZigi=1 are independent random variables, the random errors (1; ··· ; p) follow 2 T a multivariate normal distribution Np(0; σ Ip), and β = (β1; ··· ; βp) is a p-dimensional unknown vector. For simplicity, in model (1), we assume σ2 is known. Without loss of generality, let σ2 = 1.
    [Show full text]
  • A Unified Bias-Variance Decomposition for Zero-One And
    A Unified Bias-Variance Decomposition for Zero-One and Squared Loss Pedro Domingos Department of Computer Science and Engineering University of Washington Seattle, Washington 98195, U.S.A. [email protected] http://www.cs.washington.edu/homes/pedrod Abstract of a bias-variance decomposition of error: while allowing a more intensive search for a single model is liable to increase The bias-variance decomposition is a very useful and variance, averaging multiple models will often (though not widely-used tool for understanding machine-learning always) reduce it. As a result of these developments, the algorithms. It was originally developed for squared loss. In recent years, several authors have proposed bias-variance decomposition of error has become a corner- decompositions for zero-one loss, but each has signif- stone of our understanding of inductive learning. icant shortcomings. In particular, all of these decompo- sitions have only an intuitive relationship to the original Although machine-learning research has been mainly squared-loss one. In this paper, we define bias and vari- concerned with classification problems, using zero-one loss ance for an arbitrary loss function, and show that the as the main evaluation criterion, the bias-variance insight resulting decomposition specializes to the standard one was borrowed from the field of regression, where squared- for the squared-loss case, and to a close relative of Kong loss is the main criterion. As a result, several authors have and Dietterich’s (1995) one for the zero-one case. The proposed bias-variance decompositions related to zero-one same decomposition also applies to variable misclassi- loss (Kong & Dietterich, 1995; Breiman, 1996b; Kohavi & fication costs.
    [Show full text]
  • Are Loss Functions All the Same?
    Are Loss Functions All the Same? L. Rosasco∗ E. De Vito† A. Caponnetto‡ M. Piana§ A. Verri¶ September 30, 2003 Abstract In this paper we investigate the impact of choosing different loss func- tions from the viewpoint of statistical learning theory. We introduce a convexity assumption - which is met by all loss functions commonly used in the literature, and study how the bound on the estimation error changes with the loss. We also derive a general result on the minimizer of the ex- pected risk for a convex loss function in the case of classification. The main outcome of our analysis is that, for classification, the hinge loss appears to be the loss of choice. Other things being equal, the hinge loss leads to a convergence rate practically indistinguishable from the logistic loss rate and much better than the square loss rate. Furthermore, if the hypothesis space is sufficiently rich, the bounds obtained for the hinge loss are not loosened by the thresholding stage. 1 Introduction A main problem of statistical learning theory is finding necessary and sufficient condi- tions for the consistency of the Empirical Risk Minimization principle. Traditionally, the role played by the loss is marginal and the choice of which loss to use for which ∗INFM - DISI, Universit`adi Genova, Via Dodecaneso 35, 16146 Genova (I) †Dipartimento di Matematica, Universit`adi Modena, Via Campi 213/B, 41100 Modena (I), and INFN, Sezione di Genova ‡DISI, Universit`adi Genova, Via Dodecaneso 35, 16146 Genova (I) §INFM - DIMA, Universit`adi Genova, Via Dodecaneso 35, 16146 Genova (I) ¶INFM - DISI, Universit`adi Genova, Via Dodecaneso 35, 16146 Genova (I) 1 problem is usually regarded as a computational issue (Vapnik, 1995; Vapnik, 1998; Alon et al., 1993; Cristianini and Shawe Taylor, 2000).
    [Show full text]
  • Bayes Estimator Recap - Example
    Recap Bayes Risk Consistency Summary Recap Bayes Risk Consistency Summary . Last Lecture . Biostatistics 602 - Statistical Inference Lecture 16 • What is a Bayes Estimator? Evaluation of Bayes Estimator • Is a Bayes Estimator the best unbiased estimator? . • Compared to other estimators, what are advantages of Bayes Estimator? Hyun Min Kang • What is conjugate family? • What are the conjugate families of Binomial, Poisson, and Normal distribution? March 14th, 2013 Hyun Min Kang Biostatistics 602 - Lecture 16 March 14th, 2013 1 / 28 Hyun Min Kang Biostatistics 602 - Lecture 16 March 14th, 2013 2 / 28 Recap Bayes Risk Consistency Summary Recap Bayes Risk Consistency Summary . Recap - Bayes Estimator Recap - Example • θ : parameter • π(θ) : prior distribution i.i.d. • X1, , Xn Bernoulli(p) • X θ fX(x θ) : sampling distribution ··· ∼ | ∼ | • π(p) Beta(α, β) • Posterior distribution of θ x ∼ | • α Prior guess : pˆ = α+β . Joint fX(x θ)π(θ) π(θ x) = = | • Posterior distribution : π(p x) Beta( xi + α, n xi + β) | Marginal m(x) | ∼ − • Bayes estimator ∑ ∑ m(x) = f(x θ)π(θ)dθ (Bayes’ rule) | α + x x n α α + β ∫ pˆ = i = i + α + β + n n α + β + n α + β α + β + n • Bayes Estimator of θ is ∑ ∑ E(θ x) = θπ(θ x)dθ | θ Ω | ∫ ∈ Hyun Min Kang Biostatistics 602 - Lecture 16 March 14th, 2013 3 / 28 Hyun Min Kang Biostatistics 602 - Lecture 16 March 14th, 2013 4 / 28 Recap Bayes Risk Consistency Summary Recap Bayes Risk Consistency Summary . Loss Function Optimality Loss Function Let L(θ, θˆ) be a function of θ and θˆ.
    [Show full text]
  • 1 Basic Concepts 2 Loss Function and Risk
    STAT 598Y Statistical Learning Theory Instructor: Jian Zhang Lecture 1: Introduction to Supervised Learning In this lecture we formulate the basic (supervised) learning problem and introduce several key concepts including loss function, risk and error decomposition. 1 Basic Concepts We use X and Y to denote the input space and the output space, where typically we have X = Rp. A joint probability distribution on X×Y is denoted as PX,Y . Let (X, Y ) be a pair of random variables distributed according to PX,Y . We also use PX and PY |X to denote the marginal distribution of X and the conditional distribution of Y given X. Let Dn = {(x1,y1),...,(xn,yn)} be an i.i.d. random sample from PX,Y . The goal of supervised learning is to find a mapping h : X"→ Y based on Dn so that h(X) is a good approximation of Y . When Y = R the learning problem is often called regression and when Y = {0, 1} or {−1, 1} it is often called (binary) classification. The dataset Dn is often called the training set (or training data), and it is important since the distribution PX,Y is usually unknown. A learning algorithm is a procedure A which takes the training set Dn and produces a predictor hˆ = A(Dn) as the output. Typically the learning algorithm will search over a space of functions H, which we call the hypothesis space. 2 Loss Function and Risk A loss function is a mapping ! : Y×Y"→ R+ (sometimes R×R "→ R+). For example, in binary classification the 0/1 loss function !(y, p)=I(y %= p) is often used and in regression the squared error loss function !(y, p) = (y − p)2 is often used.
    [Show full text]
  • Maximum Likelihood Linear Regression
    Linear Regression via Maximization of the Likelihood Ryan P. Adams COS 324 – Elements of Machine Learning Princeton University In least squares regression, we presented the common viewpoint that our approach to supervised learning be framed in terms of a loss function that scores our predictions relative to the ground truth as determined by the training data. That is, we introduced the idea of a function ℓ yˆ, y that ( ) is bigger when our machine learning model produces an estimate yˆ that is worse relative to y. The loss function is a critical piece for turning the model-fitting problem into an optimization problem. In the case of least-squares regression, we used a squared loss: ℓ yˆ, y = yˆ y 2 . (1) ( ) ( − ) In this note we’ll discuss a probabilistic view on constructing optimization problems that fit parameters to data by turning our loss function into a likelihood. Maximizing the Likelihood An alternative view on fitting a model is to think about a probabilistic procedure that might’ve given rise to the data. This probabilistic procedure would have parameters and then we can take the approach of trying to identify which parameters would assign the highest probability to the data that was observed. When we talk about probabilistic procedures that generate data given some parameters or covariates, we are really talking about conditional probability distributions. Let’s step away from regression for a minute and just talk about how we might think about a probabilistic procedure that generates data from a Gaussian distribution with a known variance but an unknown mean.
    [Show full text]
  • Lectures on Statistics
    Lectures on Statistics William G. Faris December 1, 2003 ii Contents 1 Expectation 1 1.1 Random variables and expectation . 1 1.2 The sample mean . 3 1.3 The sample variance . 4 1.4 The central limit theorem . 5 1.5 Joint distributions of random variables . 6 1.6 Problems . 7 2 Probability 9 2.1 Events and probability . 9 2.2 The sample proportion . 10 2.3 The central limit theorem . 11 2.4 Problems . 13 3 Estimation 15 3.1 Estimating means . 15 3.2 Two population means . 17 3.3 Estimating population proportions . 17 3.4 Two population proportions . 18 3.5 Supplement: Confidence intervals . 18 3.6 Problems . 19 4 Hypothesis testing 21 4.1 Null and alternative hypothesis . 21 4.2 Hypothesis on a mean . 21 4.3 Two means . 23 4.4 Hypothesis on a proportion . 23 4.5 Two proportions . 24 4.6 Independence . 24 4.7 Power . 25 4.8 Loss . 29 4.9 Supplement: P-values . 31 4.10 Problems . 33 iii iv CONTENTS 5 Order statistics 35 5.1 Sample median and population median . 35 5.2 Comparison of sample mean and sample median . 37 5.3 The Kolmogorov-Smirnov statistic . 38 5.4 Other goodness of fit statistics . 39 5.5 Comparison with a fitted distribution . 40 5.6 Supplement: Uniform order statistics . 41 5.7 Problems . 42 6 The bootstrap 43 6.1 Bootstrap samples . 43 6.2 The ideal bootstrap estimator . 44 6.3 The Monte Carlo bootstrap estimator . 44 6.4 Supplement: Sampling from a finite population .
    [Show full text]
  • Lecture 5: Logistic Regression 1 MLE Derivation
    CSCI 5525 Machine Learning Fall 2019 Lecture 5: Logistic Regression Feb 10 2020 Lecturer: Steven Wu Scribe: Steven Wu Last lecture, we give several convex surrogate loss functions to replace the zero-one loss func- tion, which is NP-hard to optimize. Now let us look into one of the examples, logistic loss: given d parameter w and example (xi; yi) 2 R × {±1g, the logistic loss of w on example (xi; yi) is defined as | ln (1 + exp(−yiw xi)) This loss function is used in logistic regression. We will introduce the statistical model behind logistic regression, and show that the ERM problem for logistic regression is the same as the relevant maximum likelihood estimation (MLE) problem. 1 MLE Derivation For this derivation it is more convenient to have Y = f0; 1g. Note that for any label yi 2 f0; 1g, we also have the “signed” version of the label 2yi − 1 2 {−1; 1g. Recall that in general su- pervised learning setting, the learner receive examples (x1; y1);:::; (xn; yn) drawn iid from some distribution P over labeled examples. We will make the following parametric assumption on P : | yi j xi ∼ Bern(σ(w xi)) where Bern denotes the Bernoulli distribution, and σ is the logistic function defined as follows 1 exp(z) σ(z) = = 1 + exp(−z) 1 + exp(z) See Figure 1 for a visualization of the logistic function. In general, the logistic function is a useful function to convert real values into probabilities (in the range of (0; 1)). If w|x increases, then σ(w|x) also increases, and so does the probability of Y = 1.
    [Show full text]
  • Lecture 2: Estimating the Empirical Risk with Samples CS4787 — Principles of Large-Scale Machine Learning Systems
    Lecture 2: Estimating the empirical risk with samples CS4787 — Principles of Large-Scale Machine Learning Systems Review: The empirical risk. Suppose we have a dataset D = f(x1; y1); (x2; y2);:::; (xn; yn)g, where xi 2 X is an example and yi 2 Y is a label. Let h : X!Y be a hypothesized model (mapping from examples to labels) we are trying to evaluate. Let L : Y × Y ! R be a loss function which measures how different two labels are The empirical risk is n 1 X R(h) = L(h(x ); y ): n i i i=1 Most notions of error or accuracy in machine learning can be captured with an empirical risk. A simple example: measure the error rate on the dataset with the 0-1 Loss Function L(^y; y) = 0 if y^ = y and 1 if y^ 6= y: Other examples of empirical risk? We need to compute the empirical risk a lot during training, both during validation (and hyperparameter optimiza- tion) and testing, so it’s nice if we can do it fast. Question: how does computing the empirical risk scale? Three things affect the cost of computation. • The number of training examples n. The cost will certainly be proportional to n. • The cost to compute the loss function L. • The cost to evaluate the hypothesis h. Question: what if n is very large? Must we spend a large amount of time computing the empirical risk? Answer: We can approximate the empirical risk using subsampling. Idea: let Z be a random variable that takes on the value L(h(xi); yi) with probability 1=n for each i 2 f1; : : : ; ng.
    [Show full text]