arXiv subjects

Reza Hosseini

Publications and source records attributed to Reza Hosseini.

15 recordsLinked to original sources

arXiv2026

Multi-Experiment Analysis

Online controlled experiments face growing challenges from overlapping tests on shared traffic, where interactions between concurrent experiments obscure insights into feature combinations and produce effect estimates that do not correspond to any actionable launch scenario. While traffic splitting, layering, and sequential execution (non-concurrent) mitigate some of these issues, they require coordination overhead and can reduce experimentation velocity. We propose Multi-Experiment Analysis (MEA), a methodology for consistent joint estimation in the presence of arbitrary partial or full overlaps and multiple variants. MEA produces three types of estimates: (1) corrected individual treatment effects that account for the presence of overlapping experiments, (2) combined effects of launching any desired combination of variants across experiments, and (3) conditional effects of an experiment's variant given that specific variants of other experiments are launched or deramped -- all without requiring factorial pre-design or traffic restrictions. We validate the approach through comprehensive simulations confirming consistency and correct coverage. We report on production deployment at scale, illustrate the methodology through real-world use cases, and share practical lessons learned -- including system design, adoption patterns, and insights from production use.

stat.ME↗

arXiv2022

Greykite: Deploying Flexible Forecasting at Scale at LinkedIn

Forecasts help businesses allocate resources and achieve objectives. At LinkedIn, product owners use forecasts to set business targets, track outlook, and monitor health. Engineers use forecasts to efficiently provision hardware. Developing a forecasting solution to meet these needs requires accurate and interpretable forecasts on diverse time series with sub-hourly to quarterly frequencies. We present Greykite, an open-source Python library for forecasting that has been deployed on over twenty use cases at LinkedIn. Its flagship algorithm, Silverkite, provides interpretable, fast, and highly flexible univariate forecasts that capture effects such as time-varying growth and seasonality, autocorrelation, holidays, and regressors. The library enables self-serve accuracy and trust by facilitating data exploration, model configuration, execution, and interpretation. Our benchmark results show excellent out-of-the-box speed and accuracy on datasets from a variety of domains. Over the past two years, Greykite forecasts have been trusted by Finance, Engineering, and Product teams for resource planning and allocation, target setting and progress tracking, anomaly detection and root cause analysis. We expect Greykite to be useful to forecast practitioners with similar applications who need accurate, interpretable forecasts that capture complex dynamics common to time series related to human activity.

cs.LG↗

arXiv2022

Quantum Bohmian Inspired Potential to Model Non-Gaussian Events and the Application in Financial Markets

We have implemented quantum modeling mainly based on Bohmian Mechanics to study time series that contain strong coupling between their events. We firstly propose how compared to normal densities, our target time series seem to be associated with a higher number of rare events, and Gaussian statistics tend to underestimate these events' frequency drastically. To this end, we suggest that by imposing Gaussian densities to the natural processes, one will seriously neglect the existence of extreme events in many circumstances. The central question of our study concerns the consideration of the effects of these rare events in the corresponding probability densities and studying their role from the point of view of quantum measurements. To model the non-Gaussian behavior of these time-series, we utilize the multifractal random walk (MRW) approach and control the non-Gaussianity parameter $λ$ accordingly. Using the framework of quantum mechanics, we then examine the role of $λ$ in quantum potentials derived for these time series. Our Bohmian quantum analysis shows that the derived potential takes some negative values in high frequencies (its mean values), then substantially increases, and the value drops again for the rare events. We thus conclude that these events could generate a potential barrier that the system, lingering in a non-Gaussian high-frequency region, encounters, and their role becomes more prominent when it comes to transversing this barrier. In this study, as an example of the application of quantum potential outside of the micro-world, we compute the quantum potentials for the S\&P financial market time series to verify the presence of rare events in the non-Gaussian densities for this real data and remark the deviation from the Gaussian case.

q-fin.MF↗

arXiv2021

A flexible forecasting model for production systems

This paper discusses desirable properties of forecasting models in production systems. It then develops a family of models which are designed to satisfy these properties: highly customizable to capture complex patterns; accommodates a large variety of objectives; has interpretable components; produces robust results; has automatic changepoint detection for trend and seasonality; and runs fast -- making it a good choice for reliable and scalable production systems. The model allows for seasonality at various time scales, events/holidays, and change points in trend and seasonality. The volatility is fitted separately to maintain flexibility and speed and is allowed to be a function of specified features.

stat.ME↗

arXiv2020

Model selection for count timeseries with applications in forecasting number of trips in bike-sharing systems and its volatility

Forecasting the number of trips in bike-sharing systems and its volatility over time is crucial for planning and optimizing such systems. This paper develops timeseries models to forecast hourly count timeseries data, and estimate its volatility. Such models need to take into account the complex patterns over various temporal scales including hourly, daily, weekly and annual as well as the temporal correlation. To capture this complex structure, a large number of parameters are needed. Here a structural model selection approach is utilized to choose the parameters. This method explores the parameter space for a group of covariates at each step. These groups of covariate are constructed to represent a particular structure in the model. The statistical models utilized are extensions of Generalized Linear Models to timeseries data. One challenge in using such models is the explosive behavior of the simulated values. To address this issue, we develop a technique which relies on damping the simulated value, if it falls outside of an admissible interval. The admissible interval is defined using measures of variability of the left and right tails. A new definition of outliers is proposed based on these variability measures. This new definition is shown to be useful in the context of asymmetric distributions.

stat.ME↗

arXiv2019

Unbiased variance reduction in randomized experiments

This paper develops a flexible method for decreasing the variance of estimators for complex experiment effect metrics (e.g. ratio metrics) while retaining asymptotic unbiasedness. This method uses the auxiliary information about the experiment units to decrease the variance. The method can incorporate almost any arbitrary predictive model (e.g. linear regression, regularization, neural networks) to adjust the estimators. The adjustment involves some free parameters which can be optimized to achieve the smallest variance reduction given the predictive model performance. Also we approximate the achievable reduction in variance in fairly general settings mathematically. Finally, we use simulations to show the method works.

math.ST↗

arXiv2014

An objective look at obtaining the plotting positions for QQ-plots

Choosing the plotting positions for the QQ-plot has been a subject of much debate in the statistical and engineering literature. This paper looks at this problem objectively by considering three frameworks: distribution-theoretic; decision-theoretic; game-theoretic. In each framework, we derive the plotting positions and show that there are more than one legitimate solution depending on the practitioner's objective. This work clarifies the choice of the plotting positions by allowing one to easily find the mathematical equivalent of their view and choose the corresponding solution. This work also discusses approximations to the plotting positions when no closed form is available.

math.ST↗

arXiv2014

Utilizing wind in spatial covariance

This work develops a covariance function which allows for a stronger spatial correlation for pairs of points in the direction of a vector such as wind and weaker for pairs which are perpendicular to it. It derives a simple covariance function by stretching the space along the wind axes (upwind and across wind axes). It is shown that this covariance function is anisotropy in the original space and the functions is explicitly calculated.

stat.ME↗

arXiv2014

A framework for fitting sparse data

This paper develops a framework for fitting functions with domains in the Euclidean space, when data are sparse but a slow variation allows for a useful fit. We measure the variation by Lipschitz Bound (LB). Functions which admit smaller LB are considered to vary more slowly. Since most functions in practice are wiggly and do not admit a small LB, we extend this framework by approximating a wiggly function, f, by ones which admit a smaller LB and do not deviate from f by more than a specified Bound Deviation (BD). In fact for any positive LB, one can find such a BD, thus defining a trade-off function (LB-BD function) between the variation measure (LB) and the deviation measure (BD). We show that the LB-BD function satisfies nice properties: it is non-increasing and convex. We also present a method to obtain it using convex optimization. For a function with given LB and BD, we find the optimal fit and present deterministic bounds for the prediction error of various methods. Given the LB-BD function, we discuss picking an appropriate LB-BD pair for fitting and calculating the prediction errors. The developed methods can naturally accommodate an extra assumption of periodicity to obtain better prediction errors. Finally we present the application of this framework to air pollution data with sparse observations over time.

stat.ME↗

arXiv2010

Conditional information and definition of neighbor in categorical random fields

We show that the definition of neighbor in Markov random fields as defined by Besag (1974) when the joint distribution of the sites is not positive is not well-defined. In a random field with finite number of sites we study the conditions under which giving the value at extra sites will change the belief of an agent about one site. Also the conditions under which the information from some sites is equivalent to giving the value at all other sites is studied. These concepts provide an alternative to the concept of neighbor for general case where the positivity condition of the joint does not hold.

math.ST↗

arXiv2010

Closure of Random samples

In this paper we show that the closure of a random sample for a k-dimensional random vector is almost surely a deterministic set of all heavy points of the distribution. A heavy point is defined to be a point for which all its neighborhoods have positive probability.

math.PR↗

arXiv2010

Approximating quantiles in very large datasets

Very large datasets are often encountered in climatology, either from a multiplicity of observations over time and space or outputs from deterministic models (sometimes in petabytes= 1 million gigabytes). Loading a large data vector and sorting it, is impossible sometimes due to memory limitations or computing power. We show that a proposed algorithm to approximating the median, "the median of the median" performs poorly. Instead we develop an algorithm to approximate quantiles of very large datasets which works by partitioning the data or use existing partitions (possibly of non-equal size). We show the deterministic precision of this algorithm and how it can be adjusted to get customized precisions.

stat.CO↗

arXiv2010

Divergence of sample quantiles

We show that the left (right) sample quantile tends to the left (right) distribution quantile at p in [0,1], if the left and right quantiles are identical at p. We show that the sample quantiles diverge almost surely otherwise. The latter can be considered as a generalization of the well-known result that the sum of a random sample of a fair coin with 1 denoting heads and -1 denoting tails is 0 infinitely often. In the case that the sample quantiles do not converge we show that the limsup is the right quantile and the liminf is the left quantile.

math.ST↗

arXiv2010

Quantiles Equivariance

It is widely claimed that the quantile function is equivariant under increasing transformations. We show by a counterexample that this is not true (even for strictly increasing transformations). However, we show that the quantile function is equivariant under left continuous increasing transformations. We also provide an equivariance relation for continuous decreasing transformations. In the case that the transformation is not continuous, we show that while the transformed quantile at p can be arbitrarily far from the quantile of the transformed at p (in terms of absolute difference), the probability mass between the two is zero. We also show by an example that weighted definition of the median is not equivariant under even strictly increasing continuous transformations.

math.ST↗

arXiv2010

Quantiles symmetry

This paper finds a symmetry relation (between quantiles of a random variable and its negative) that is intuitively appealing. We show this symmetry is quite useful in finding new relations for quantiles, in particular an equivariance property for quantiles under continuous decreasing transformations.

math.ST↗