arXiv subjects

Robert Tibshirani

Publications and source records attributed to Robert Tibshirani.

At least 91 records · Page 5Linked to original sources

arXiv2014

Regularisation Paths for Conditional Logistic Regression: the clogitL1 package

We apply the cyclic coordinate descent algorithm of Friedman, Hastie and Tibshirani (2010) to the fitting of a conditional logistic regression model with lasso ($\ell_1$) and elastic net penalties. The sequential strong rules of Tibshirani et al (2012) are also used in the algorithm and it is shown that these offer a considerable speed up over the standard coordinate descent algorithm with warm starts. Once implemented, the algorithm is used in simulation studies to compare the variable selection and prediction performance of the conditional logistic regression model against that of its unconditional (standard) counterpart. We find that the conditional model performs admirably on datasets drawn from a suitable conditional distribution, outperforming its unconditional counterpart at variable selection. The conditional model is also fit to a small real world dataset, demonstrating how we obtain regularisation paths for the parameters of the model and how we apply cross validation for this method where natural unconditional prediction rules are hard to come by.

stat.ME↗

arXiv2014

A Study of Error Variance Estimation in Lasso Regression

Variance estimation in the linear model when $p > n$ is a difficult problem. Standard least squares estimation techniques do not apply. Several variance estimators have been proposed in the literature, all with accompanying asymptotic results proving consistency and asymptotic normality under a variety of assumptions. It is found, however, that most of these estimators suffer large biases in finite samples when true underlying signals become less sparse with larger per element signal strength. One estimator seems to be largely neglected in the literature: a residual sum of squares based estimator using Lasso coefficients with regularisation parameter selected adaptively (via cross-validation). In this paper, we review several variance estimators and perform a reasonably extensive simulation study in an attempt to compare their finite sample performance. It would seem from the results that variance estimators with adaptively chosen regularisation parameters perform admirably over a broad range of sparsity and signal strength settings. Finally, some intial theoretical analyses pertaining to these types of estimators are proposed and developed.

stat.ME↗

arXiv2014

Comment on "Detecting Novel Associations In Large Data Sets" by Reshef Et Al, Science Dec 16, 2011

The proposal of Reshef et al. (2011) is an interesting new approach for discovering non-linear dependencies among pairs of measurements in exploratory data mining. However, it has a potentially serious drawback. The authors laud the fact that MIC has no preference for some alternatives over others, but as the authors know, there is no free lunch in Statistics: tests which strive to have high power against all alternatives can have low power in many important situations. To investigate this, we ran simulations to compare the power of MIC to that of standard Pearson correlation and distance correlation (dcor). We simulated pairs of variables with different relationships (most of which were considered by the Reshef et. al.), but with varying levels of noise added. To determine proper cutoffs for testing the independence hypothesis, we simulated independent data with the appropriate marginals. As one can see from the Figure, MIC has lower power than dcor, in every case except the somewhat pathological high-frequency sine wave. MIC is sometimes less powerful than Pearson correlation as well, the linear case being particularly worrisome.

stat.ME↗

arXiv2014

Collaborative Regression

We consider the scenario where one observes an outcome variable and sets of features from multiple assays, all measured on the same set of samples. One approach that has been proposed for dealing with this type of data is ``sparse multiple canonical correlation analysis'' (sparse mCCA). All of the current sparse mCCA techniques are biconvex and thus have no guarantees about reaching a global optimum. We propose a method for performing sparse supervised canonical correlation analysis (sparse sCCA), a specific case of sparse mCCA when one of the datasets is a vector. Our proposal for sparse sCCA is convex and thus does not face the same difficulties as the other methods. We derive efficient algorithms for this problem, and illustrate their use on simulated and real data.

q-bio.QM↗

arXiv2013

A Component Lasso

We propose a new sparse regression method called the component lasso, based on a simple idea. The method uses the connected-components structure of the sample covariance matrix to split the problem into smaller ones. It then solves the subproblems separately, obtaining a coefficient vector for each one. Then, it uses non-negative least squares to recombine the different vectors into a single solution. This step is useful in selecting and reweighting components that are correlated with the response. Simulated and real data examples show that the component lasso can outperform standard regression methods such as the lasso and elastic net, achieving a lower mean squared error as well as better support recovery.

stat.ML↗

arXiv2013

An Investigation of Methods for Handling Missing Data with Penalized Regression

We investigate methods for penalized regression in the presence of missing observations. This paper introduces a method for estimating the parameters which compensates for the missing observations. We first, derive an unbiased estimator of the objective function with respect to the missing data and then, modify the criterion to ensure convexity. Finally, we extend our approach to a family of models that embraces the mean imputation method. These approaches are compared to the mean imputation method, one of the simplest methods for dealing with missing observations problem, via simulations. We also investigate the problem of making predictions when there are missing values in the test set.

stat.AP↗

arXiv2013

Sensitivity Analysis for Inference with Partially Identifiable Covariance Matrices

In some multivariate problems with missing data, pairs of variables exist that are never observed together. For example, some modern biological tools can produce data of this form. As a result of this structure, the covariance matrix is only partially identifiable, and point estimation requires that identifying assumptions be made. These assumptions can introduce an unknown and potentially large bias into the inference. This paper presents a method based on semidefinite programming for automatically quantifying this potential bias by computing the range of possible equal-likelihood inferred values for convex functions of the covariance matrix. We focus on the bias of missing value imputation via conditional expectation and show that our method can give an accurate assessment of the true error in cases where estimates based on sampling uncertainty alone are overly optimistic.

stat.ME↗

arXiv2013

Adaptive testing for the graphical lasso

We consider tests of significance in the setting of the graphical lasso for inverse covariance matrix estimation. We propose a simple test statistic based on a subsequence of the knots in the graphical lasso path. We show that this statistic has an exponential asymptotic null distribution, under the null hypothesis that the model contains the true connected components. Though the null distribution is asymptotic, we show through simulation that it provides a close approximation to the true distribution at reasonable sample sizes. Thus the test provides a simple, tractable test for the significance of new edges as they are introduced into the model. Finally, we show connections between our results and other results for regularized regression, as well as extensions of our results to other correlation matrix based methods like single-linkage clustering.

math.ST↗

arXiv2013

A lasso for hierarchical interactions

We add a set of convex constraints to the lasso to produce sparse interaction models that honor the hierarchy restriction that an interaction only be included in a model if one or both variables are marginally important. We give a precise characterization of the effect of this hierarchy constraint, prove that hierarchy holds with probability one and derive an unbiased estimate for the degrees of freedom of our estimator. A bound on this estimate reveals the amount of fitting "saved" by the hierarchy constraint. We distinguish between parameter sparsity - the number of nonzero coefficients - and practical sparsity - the number of raw variables one must measure to make a new prediction. Hierarchy focuses on the latter, which is more closely tied to important data collection concerns such as cost, time and effort. We develop an algorithm, available in the R package hierNet, and perform an empirical study of our method.

stat.ME↗

arXiv2013

False Variable Selection Rates in Regression

There has been recent interest in extending the ideas of False Discovery Rates (FDR) to variable selection in regression settings. Traditionally the FDR in these settings has been defined in terms of the coefficients of the full regression model. Recent papers have struggled with controlling this quantity when the predictors are correlated. This paper shows that this full model definition of FDR suffers from unintuitive and potentially undesirable behavior in the presence of correlated predictors. We propose a new false selection error criterion, the False Variable Rate (FVR), that avoids these problems and behaves in a more intuitive manner. We discuss the behavior of this criterion and how it compares with the traditional FDR, as well as presenting guidelines for determining which is appropriate in a particular setting. Finally, we present a simple estimation procedure for FVR in stepwise variable selection. We analyze the performance of this estimator and draw connections to recent estimators in the literature.

stat.ME↗

arXiv2012

A Simple Method for Detecting Interactions between a Treatment and a Large Number of Covariates

We consider a setting in which we have a treatment and a large number of covariates for a set of observations, and wish to model their relationship with an outcome of interest. We propose a simple method for modeling interactions between the treatment and covariates. The idea is to modify the covariate in a simple way, and then fit a standard model using the modified covariates and no main effects. We show that coupled with an efficiency augmentation procedure, this method produces valid inferences in a variety of settings. It can be useful for personalized medicine: determining from a large set of biomarkers the subset of patients that can potentially benefit from a treatment. We apply the method to both simulated datasets and gene expression studies of cancer. The modified data can be used for other purposes, for example large scale hypothesis testing for determining which of a set of covariates interact with a treatment variable.

stat.ME↗

arXiv2012

A Permutation Approach to Testing Interactions in Many Dimensions

To date, testing interactions in high dimensions has been a challenging task. Existing methods often have issues with sensitivity to modeling assumptions and heavily asymptotic nominal p-values. To help alleviate these issues, we propose a permutation-based method for testing marginal interactions with a binary response. Our method searches for pairwise correlations which differ between classes. In this manuscript, we compare our method on real and simulated data to the standard approach of running many pairwise logistic models. On simulated data our method finds more significant interactions at a lower false discovery rate (especially in the presence of main effects). On real genomic data, although there is no gold standard, our method finds apparent signal and tells a believable story, while logistic regression does not. We also give asymptotic consistency results under not too restrictive assumptions.

stat.ML↗

arXiv2012

Prototype selection for interpretable classification

Prototype methods seek a minimal subset of samples that can serve as a distillation or condensed view of a data set. As the size of modern data sets grows, being able to present a domain specialist with a short list of "representative" samples chosen from the data set is of increasing interpretative value. While much recent statistical research has been focused on producing sparse-in-the-variables methods, this paper aims at achieving sparsity in the samples. We discuss a method for selecting prototypes in the classification setting (in which the samples fall into known discrete categories). Our method of focus is derived from three basic properties that we believe a good prototype set should satisfy. This intuition is translated into a set cover optimization problem, which we solve approximately using standard approaches. While prototype selection is usually viewed as purely a means toward building an efficient classifier, in this paper we emphasize the inherent value of having a set of prototypical elements. That said, by using the nearest-neighbor rule on the set of prototypes, we can of course discuss our method as a classifier as well.

stat.AP↗

arXiv2010

Strong rules for discarding predictors in lasso-type problems

We consider rules for discarding predictors in lasso regression and related problems, for computational efficiency. El Ghaoui et al (2010) propose "SAFE" rules that guarantee that a coefficient will be zero in the solution, based on the inner products of each predictor with the outcome. In this paper we propose strong rules that are not foolproof but rarely fail in practice. These can be complemented with simple checks of the Karush- Kuhn-Tucker (KKT) conditions to provide safe rules that offer substantial speed and space savings in a variety of statistical convex optimization problems.

math.ST↗

arXiv2010

Transposable regularized covariance models with an application to missing data imputation

Missing data estimation is an important challenge with high-dimensional data arranged in the form of a matrix. Typically this data matrix is transposable, meaning that either the rows, columns or both can be treated as features. To model transposable data, we present a modification of the matrix-variate normal, the mean-restricted matrix-variate normal, in which the rows and columns each have a separate mean vector and covariance matrix. By placing additive penalties on the inverse covariance matrices of the rows and columns, these so-called transposable regularized covariance models allow for maximum likelihood estimation of the mean and nonsingular covariance matrices. Using these models, we formulate EM-type algorithms for missing data imputation in both the multivariate and transposable frameworks. We present theoretical results exploiting the structure of our transposable models that allow these models and imputation methods to be applied to high-dimensional data. Simulations and results on microarray data and the Netflix data show that these imputation techniques often outperform existing methods and offer a greater degree of flexibility.

stat.AP↗

arXiv2010

Bayesian Gene Set Analysis

Gene expression microarray technologies provide the simultaneous measurements of a large number of genes. Typical analyses of such data focus on the individual genes, but recent work has demonstrated that evaluating changes in expression across predefined sets of genes often increases statistical power and produces more robust results. We introduce a new methodology for identifying gene sets that are differentially expressed under varying experimental conditions. Our approach uses a hierarchical Bayesian framework where a hyperparameter measures the significance of each gene set. Using simulated data, we compare our proposed method to alternative approaches, such as Gene Set Enrichment Analysis (GSEA) and Gene Set Analysis (GSA). Our approach provides the best overall performance. We also discuss the application of our method to experimental data based on p53 mutation status.

stat.AP↗

arXiv2010

Inference with Transposable Data: Modeling the Effects of Row and Column Correlations

We consider the problem of large-scale inference on the row or column variables of data in the form of a matrix. Often this data is transposable, meaning that both the row variables and column variables are of potential interest. An example of this scenario is detecting significant genes in microarrays when the samples or arrays may be dependent due to underlying relationships. We study the effect of both row and column correlations on commonly used test-statistics, null distributions, and multiple testing procedures, by explicitly modeling the covariances with the matrix-variate normal distribution. Using this model, we give both theoretical and simulation results revealing the problems associated with using standard statistical methodology on transposable data. We solve these problems by estimating the row and column covariances simultaneously, with transposable regularized covariance models, and de-correlating or sphering the data as a pre-processing step. Under reasonable assumptions, our method gives test statistics that follow the scaled theoretical null distribution and are approximately independent. Simulations based on various models with structured and observed covariances from real microarray data reveal that our method offers substantial improvements in two areas: 1) increased statistical power and 2) correct estimation of false discovery rates.

stat.ME↗

arXiv2009

A bias correction for the minimum error rate in cross-validation

Tuning parameters in supervised learning problems are often estimated by cross-validation. The minimum value of the cross-validation error can be biased downward as an estimate of the test error at that same value of the tuning parameter. We propose a simple method for the estimation of this bias that uses information from the cross-validation process. As a result, it requires essentially no additional computation. We apply our bias estimate to a number of popular classifiers in various settings, and examine its performance.

stat.AP↗