Treed Gaussian Process Research Articles

Treed Gaussian processes for animal movement modeling.

Wildlife telemetry data may be used to answer a diverse range of questions relevant to wildlife ecology and management. One challenge to modeling telemetry data is that animal movement often varies greatly in pattern over time, and current continuous-time modeling approaches to handle such nonstationarity require bespoke and often complex models that may pose barriers to practitioner implementation. We demonstrate a novel application of treed Gaussian process (TGP) modeling, a Bayesian machine learning approach that automatically captures the nonstationarity and abrupt transitions present in animal movement. The machine learning formulation of TGPs enables modeling to be nearly automated, while their Bayesian formulation allows for the derivation of movement descriptors with associated uncertainty measures. We demonstrate the use of an existing R package to implement TGPs using the familiar Markov chain Monte Carlo algorithm. We then use estimated movement trajectories to derive movement descriptors that can be compared across individuals and populations. We applied the TGP model to a case study of lesser prairie-chickens (Tympanuchus pallidicinctus) to demonstrate the benefits of TGP modeling and compared distance traveled and residence times across lesser prairie-chicken individuals and populations. For broad usability, we outline all steps necessary for practitioners to specify relevant movement descriptors (e.g., turn angles, speed, contact points) and apply TGP modeling and trajectory comparison to their own telemetry datasets. Combining the predictive power of machine learning and the statistical inference of Bayesian methods to model movement trajectories allows for the estimation of statistically comparable movement descriptors from telemetry studies. Our use of an accessible R package allows practitioners to model trajectories and estimate movement descriptors, facilitating the use of telemetry data to answer applied management questions.

Treed Gaussian Process Regression for Solving Offline Data-Driven Continuous Multiobjective Optimization Problems.

For offline data-driven multiobjective optimization problems (MOPs), no new data is available during the optimization process. Approximation models (or surrogates) are first built using the provided offline data, and an optimizer, for example, a multiobjective evolutionary algorithm, can then be utilized to find Pareto optimal solutions to the problem with surrogates as objective functions. In contrast to online data-driven MOPs, these surrogates cannot be updated with new data and, hence, the approximation accuracy cannot be improved by considering new data during the optimization process. Gaussian process regression (GPR) models are widely used as surrogates because of their ability to provide uncertainty information. However, building GPRs becomes computationally expensive when the size of the dataset is large. Using sparse GPRs reduces the computational cost of building the surrogates. However, sparse GPRs are not tailored to solve offline data-driven MOPs, where good accuracy of the surrogates is needed near Pareto optimal solutions. Treed GPR (TGPR-MO) surrogates for offline data-driven MOPs with continuous decision variables are proposed in this paper. The proposed surrogates first split the decision space into subregions using regression trees and build GPRs sequentially in regions close to Pareto optimal solutions in the decision space to accurately approximate tradeoffs between the objective functions. TGPR-MO surrogates are computationally inexpensive because GPRs are built only in a smaller region of the decision space utilizing a subset of the data. The TGPR-MO surrogates were tested on distance-based visualizable problems with various data sizes, sampling strategies, numbers of objective functions, and decision variables. Experimental results showed that the TGPR-MO surrogates are computationally cheaper and can handle datasets of large size. Furthermore, TGPR-MO surrogates produced solutions closer to Pareto optimal solutions compared to full GPRs and sparse GPRs.

Modeling the R-ratio and hadronic contributions to g-2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$g-2$$\\end{document} with a Treed Gaussian process

The BNL and FNAL measurements of the anomalous magnetic moment of the muon disagree with the Standard Model (SM) prediction by more than 4σ\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$4\\sigma $$\\end{document}. The hadronic vacuum polarization (HVP) contributions are the dominant source of uncertainty in the SM prediction. There are, however, tensions between different estimates of the HVP contributions, including data-driven estimates based on measurements of the R-ratio. To investigate that tension, we modeled the unknown R-ratio as a function of CM energy with a treed Gaussian process (TGP). This is a principled and general method grounded in data-science that allows complete uncertainty quantification and automatically balances over- and under-fitting to noisy data. Our tool yields exploratory results are similar to previous ones and we find no indication that the R-ratio was previously mismodeled. Whilst we advance some aspects of modeling the R-ratio and develop new tools for doing so, a competitive estimate of the HVP contributions requires domain-specific expertise and a carefully curated database of measurements (github, https://github.com/qiao688/TGP_for_g-2).

Uncertainty analysis of envelope retrofits for existing residential buildings in underdeveloped areas: A case study of Daokou, China

Current envelope-related uncertainty analysis (UA) focuses mostly on the physical properties of materials. However, the uncertainties of retrofit strategy selection, building element selection and construction quality are especially significant in underdeveloped areas due to their low economic and technical levels. To obtain more reliable energy analysis results and to correctly support decision-makers, a new UA method combining an orthogonal experiment (OE) and Latin hypercube sampling (LHS) is proposed to investigate the uncertain outputs caused by the three mentioned aspects. Furthermore, important factors affecting the uncertain outputs are screened by two global sensitivity analysis (GSA) methods: the standard regression coefficient (SRC) method and the treed Gaussian process (TGP) method. The UA results show that considering the uncertain inputs, the envelope energy-saving rates fluctuate from 46 %∼65 %, 70 %∼92 %, 48 %∼67 %, and 71 %∼90 % separately for the four typical retrofit scenes. The retrofit scene with the best energy-saving effect is always accompanied by the highest uncertainty. The GSA results show that the top three sensitive factors account for approximately 80 ∼ 100 % of the output variance. Infiltration and wall insulation thickness are the two most important factors for energy consumption in most cases. The results of the SRC and TGP methods are in general agreement, while TGP has the advantage of distinguishing the nonlinear relationship between glazing type and cooling energy. Therefore, it is advisable to apply design schemes that can make a trade-off between energy savings and robustness when performing envelope retrofits in underdeveloped areas.

Global sensitivity analysis and stochastic optimization of multi-energy complementary distributed energy system considering multiple uncertainties

Multiple uncertainties exist in the planning and design of multi-energy complementary distributed energy system (DES). In order to solve this problem, a more complete framework of two-stage stochastic programming (TSP) approach for optimal multi-energy complementary DES planning and design under multiple uncertainties is proposed. The developed TSP model is formulated as a mixed integer linear programming, which is intended to minimize the equivalent annual cost (EAC) of the multi-energy complementary DES. Probability scenarios of key uncertain parameters selected by treed Gaussian process method for the TSP model are generated and reduced using discrete approximations of probability distributions method and random vector sampling method. The developed method framework is applied to a case study high-rise office building in planning stage to illustrate the TSP model's output. And multiple evaluation indicators including configuration, economy, energy and environment are used to compare and analyze the optimal schemes obtained by deterministic model and TSP model under different building load scenarios. The results show that of the 26 uncertain parameters considered only 7 are selected as the key parameters for the TSP model in this paper. EAC of the optimal multi-energy complementary DES may be overestimated if without considering key uncertain parameters. Thus, the deterministic optimization practices and the cost estimates resulting from them can be considered unreliable. Moreover, the multi-energy complementary DES would want to increase CO2 emission to achieve better economic performance. Finally, the energy share distribution of all scenarios in the stochastic model provides more reliable and more accurate information for decision-makers.

A localized ensemble of approximate Gaussian processes for fast sequential emulation

More attention has been given to the computational cost associated with the fitting of an emulator. Substantially less attention is given to the computational cost of using that emulator for prediction. This is primarily because the cost of fitting an emulator is usually far greater than that of obtaining a single prediction, and predictions can often be obtained in parallel. In many settings, especially those requiring Markov Chain Monte Carlo, predictions may arrive sequentially and parallelization is not possible. In this case, using an emulator procedure which can produce accurate predictions efficiently can lead to substantial time savings in practice. In this paper, we propose a global model approximate Gaussian process framework via extension of a popular local approximate Gaussian process (laGP) framework. Our proposed emulator can be viewed as a treed Gaussian process where the leaf nodes are laGP models, and the tree structure is learned greedily as a function of the prediction stream. The suggested method (called leapGP) has interpretable tuning parameters which control the time‐memory trade‐off. One reasonable choice of settings leads to an emulator with a training cost and makes predictions rapidly with an asymptotic amortized cost of .

Uncertainty quantification and sensitivity analysis of energy consumption in substation buildings at the planning stage

ABSTRACT This study is the first to conduct a global sensitivity analysis to identify the crucial variables that have an impact on the energy consumption of substations. The peak cooling and heating energy consumption, as well as the annual cooling and heating energy consumption of a substation in Shandong, are all simulated basing the Monte Carlo method. The simulation outputs are discussed by uncertainty analysis to obtain more accurate energy consumption thresholds. Subsequently, the treed Gaussian process and the standardized rank regression coefficient are used to perform a global sensitivity analysis of the simulation outputs. The results of the two global sensitivity analyses are practically the same, demonstrating that robustness can be ensured by simultaneously using two methods based on different theories. In addition, this study provides an efficient method for the energy-saving retrofitting of the existing substation and the energy-saving design of green substations in the future.

Global Sensitivity Analysis of Key Parameters in the APSIMX-Sugarcane Model to Evaluate Nitrate Balance via Treed Gaussian Process

Difficulties in direct monitoring of nitrate balance in agricultural fields reveal the importance of modeling and quantifying the affecting parameters on nitrate balance. We constructed meta-models for APSIMX-Sugarcane using the treed gaussian process and conducted a global sensitivity analysis for nitrate uptake and leaching under three conditions: (1) bare land (BL) to examine the influence of soil hydraulic characteristics, (2) N-free treatment under radiation use efficiency (RUE) ranges (i) 1.2–1.8 [N-free(a)] and (ii) 1.8–2.5 [N-free(b)], and (3) urea conditions to examine the influence of plant growth. Generated meta-models showed good accuracy (for all conditions: R2 > 0.70; NRMSE < 16%; AI > 0.90). The most influential parameters (sensitivity indices ≥ 0.02) were as follows: for leached NO3−N in BL: the parameter rerated to saturated flow-proportion of water between saturation and field capacity (SWCON) of all soil layers; for NO3− uptake and leached NO3−N in N-free(a) and urea: RUE of the phenological stage (PS) 3 (RUE3) and 4, tt_emerg_to_begcane, green_leaf_no, and y_n_conc_crit_leaf of PS 4 (NCL4); in N-free(b): RUE3, NCL4, and SWCON of soil layers 0–15 cm; 15–30 cm, which confirmed that influential parameters were depended on N-stress. The outcomes of this study are useful for enhancing the accuracy and efficiency of crop modeling.

Sensitivity analysis of environmental impact of milk powder spray drying system based on intelligent algorithm

The increasing market demand for milk powder has not only promoted the production capacity of milk powder, but also increased the impact on the environment. Therefore, it is very important to study the relationship between the environmental impact of milk powder spray drying (MPSD) system and system-related parameters and identify the key parameters to improve the efficiency of the sustainable improvement of the system. Treed Gaussian Process (TGP) and Standardized Regression Coefficients (SRC)methods are used to analyze the sensitivity of the system to environmental impacts. The results show that the inlet air temperature of the drying tower has the greatest impact on the environment of the system, accounting for about 82%, followed by the atomization pressure and the feed pump speed, accounting for about 9% and 8% respectively. Moreover, not only the environmental performance of the system should be improved, but also the quality of milk powder should be guaranteed when optimizing the parameters such as the inlet air temperature of drying tower. This study can help the manufacturers of milk powder and related equipment to determine the priority of improving the system from the perspective of environmental protection.

Impacts of Courtyard Envelope Design on Energy Performance in the Hot Summer–Cold Winter Region of China

This paper aims to investigate the effects of courtyard envelope design on the energy performance of office buildings in the hot summer–cold winter region of China. Two types of courtyard buildings were simulated with 200 energy models by changing the following variables: window–wall ratio (south, north, east, and west walls), window U-value, wall U-value, solar heat gain coefficient, and orientation. The treed Gaussian process (TGP) sensitivity analysis method was employed to quantify the contributions of parameters related to courtyard design that result in the changes of annual cooling, heating, lighting, and total energy consumption. The results show that the courtyard envelope design has a significant effect on the energy consumption in this case study. The most influential factor affecting annual cooling energy consumption and the annual total energy consumption is the orientation, which is responsible for 37% and 81%, respectively, in two layouts to annual cooling energy use. The corresponding proportion increased to 45% and 86% in terms of the total energy consumption. The most influential factor influencing annual heating energy use is the window U-value, which explained nearly 60% of the changes to the heating energy use. The effect of the solar heat gain coefficient in two kinds of layouts is as high as 82% and 79% for lighting energy use in this case study. Due to the different courtyard forms, the magnitude of the effect of the parameters on energy consumption and the main trend of the effect is slightly different.

Coupling effect of building design variables on building energy performance

In this paper, the coupling effect of building design variables on building energy performance was studied by taking office in hot summer and cold winter zone as an example. The Latin hypercube method was introduced to sample the architectural design features under different combinations of 10 variables, and simultaneously, the simulations on energy consumption were conducted through Energyplus. The potential interaction effect between design variables were analysed by using treed Gaussian process. Meanwhile, the coupling effect of these variables was further analysed quantitatively by using interaction in linear regression. The study showed that the coupling effect of the design variables and their impact on energy consumption trends could not be revealed by ordinary linear regression. Besides, the influence of interactions between design variables on energy consumption needs to be taken into account. The coupling relationships were found amongst the windows' heat transfer coefficient, the window-to-wall ratio, and the air infiltration. Therefore, these parameters should be selected properly based on the features of energy consumption and coupling variables. Furthermore, it was recommended to consider summer heat insulation/dissipation when the window-to-wall ratio is large in this district, instead of strengthening heat preservation.

Overall and thermal comfort under different temperature, noise, and vibration exposures

Public transports like the bus and subway inherently experience noise, vibration, and temperature variations that are different from building environment. Each of them can influence passengers' comfort, but little is known about their combined effects, especially how they affect thermal comfort. This paper presents experimental results from a series of human subject tests under different noises, vibrations, and temperatures. 32subjects' subjective perception and physiological response were collected under three temperatures (22.5, 25.5, 28.5℃), five noise levels (55, 60, 65, 70, 75dB(A)), and five vibrating accelerations (0, 0.2, 0.4, 0.6, 0.8m/s2 ). We also varied the noise and vibration spectrums to simulate the bus and subway environments. In total, 48 195-min and 192 115-min laboratory tests were conducted. By using significance tests (paired t tests and two-way ANOVA tests) and sensitivity analysis (Treed Gaussian Process), the results show that temperature, noise, and vibration exposures can significantly affect subjects' overall satisfaction. More interestingly, high noise and vibration levels can cause warmer thermal sensations. A change in the noise of 20dB(A) or vibration of 0.6m/s2 is equivalent to an ambient temperature change of 0.6 °C. We also observed higher heart rates and metabolic heat production at higher levels of noise and vibrating accelerators. Based on the test results, regression models were developed to describe the combined effects of temperature, noise, and vibration on subjects' overall comfort perception and thermal neutral temperature. They can serve as references for the design and operation of public transport environments.

Uncertainty and global sensitivity analysis on thermal performances of pipe-embedded building envelope in the heating season

Pipe-embedded building envelope, which is of particular interest to architects and engineers due to its excellent performances and invisibility characteristic, has been regarded as a potential technology for future buildings. Understanding the impacts of multiple uncertainties associated with design, construction, and operational control on its thermal performances is critical to the development of this burgeoning active insulation technology. Aiming at this goal, an in-depth uncertainty analysis (UA) and global sensitivity analysis (GSA) study was numerically conducted to investigate the influence mechanism of 12 different variables on 5 performance indicators in winter conditions. The UA results indicated that a dual-effect, i.e., better indoor thermal comfort and “zero” or even “negative” thermal load of the exterior walls, could be achieved only when relevant variables were properly selected. Due to the increase in the quantities and types of input variables, as well as the internal heat transfer process became more complicated, the importance rankings of variables obtained by two different GSA methods became different. Therefore, the treed Gaussian process method was proven to be more effective in identifying the uncertain variables. The GSA results stated that heat-source temperature, indoor set-point, charging duration, and thermal conductivity of pipe-embedded layer were the four most significant variables. For heat-source temperature and indoor set-point, there always existed an obvious mutual restriction relationship between them among all indicators except exterior surface heating loss. Meanwhile, the suggested value of the charging duration was no less than 6 h, and the optimal range of the thermal conductivity of pipe-embedded layer given by GSA was 0.5–2.75 W/m·℃. Under the blocking effect of the thermal barrier, the uncertainty of the climate zone on interior surface heating load was greatly reduced, and the climate zone was no longer a key variable affecting the cumulative subcooling duration. Besides, it was proved that the pipe spacing had a great influence on the heat accumulation inside the pipe-embedded layer, pipe location had a slight influence on interior surface heating load and exterior surface heating loss, and the pipe diameter did not influence thermal performances. From the perspective of energy density, the optimal range of the pipe spacing was 100–250 mm. Overall, this study highlighted the positive effect of the pipe-embedded system on the development of new-built and existing buildings towards zero-carbon targets, and also provided useful guidance for its further applications and investigations.

Treed gaussian process for manufacturing imperfection identification of pultruded GFRP thin-walled profile

The process of manufacturing pultruded FRP (Fiber Reinforced Polymers) profiles involves unavoidable imperfections that affect their structural performances. This is is even more relevant for the stability of axially loaded slender elements, due to the importance of imperfections and notches to initiate the buckling phenomenon. Thus, they become a predominant factor for the design of lightweight FRP beam-like structures. A Bayesian approach is proposed to estimate the presence and location of manufacturing imperfections in pultruded GFRPs (Glass Fiber Reinforced Polymers) profiles. Specifically, the Treed Gaussian Process (TGP) procedure is applied. This approach combines regression Gaussian Processes (GP) and Bayesian-based Recursive Partitioning. The experimental and numerical modal shapes of wide flange pultruded profile were investigated. The experimental data were compared with the numerical results of several Finite Element Models (FEM) characterised by different crack sizes.

An Improved Treed Gaussian Process

Many black box functions and datasets have regions of different variability. Models such as the Gaussian process may fall short in giving the best representation of these complex functions. One successful approach for modeling this type of nonstationarity is the Treed Gaussian process [1], which extended the Gaussian process by dividing the input space into different regions using a binary tree algorithm. Each region became its own Gaussian process. This iterative inference process formed many different trees and thus, many different Gaussian processes. In the end these were combined to get a posterior predictive distribution at each point. The idea was that when the iterations were combined, smoothing would take place for the surface of the predicted points near tree boundaries. We introduce the Improved Treed Gaussian process, which divides the input space into a single main binary tree where the different tree regions have different variability. The parameters for the Gaussian process for each tree region are then determined. These parameters are then smoothed at the region boundaries. This smoothing leads to a set of parameters for each point in the input space that specify the covariance matrix used to predict the point. The advantage is that the prediction and actual errors are estimated better since the standard deviation and range parameters of each point are related to the variation of the region it is in. Further, smoothing between regions is better since each point prediction uses its parameters over the whole input space. Examples are given in this paper which show these advantages for lower-dimensional problems.

Heat Transfer Performance of Vertical Tube Falling Film Juice Evaporator and a Sensitivity Analysis

Abstract Fresh fruit juice has a short storage period and it usually needs to be concentrated for shelf-life extension. Evaporation of fruit juice is an attractive method and vertical tube falling film evaporator is one of good choice. The main problem in the design of the evaporator is the reliable estimation of the heat transfer coefficient for all effects. In this paper, the effects of feed flow rate, evaporation temperature, temperature difference, and juice soluble solid content on heat transfer coefficient of a vertical falling film evaporator were investigated. The Bayesian treed Gaussian process model was used to identify the key variables affecting the heat transfer coefficient, which showed that evaporation temperature had the greatest influence on the sensitivity of heat transfer coefficient. It also demonstrated that complex interactions existed between the four operating parameters had a significant effect on the overall heat transfer coefficient.

Thermal performance of a novel ultrasonic evaporator based on machine learning algorithms

Ultrasound is a promising method to enhance heat transfer in industrial evaporators. However, there are very limited studies on this topic. This paper explores thermal performance of a novel ultrasonic evaporator based on machine learning methods. The results indicate that the overall heat transfer coefficients can be increased by around 15–20% after adding ultrasound due to acoustic cavitation and acoustic streaming. Two machine learning-based global sensitivity analysis (treed Gaussian Process and polynomial chaos expansion) methods are used to identify important variables influencing overall heat transfer coefficients in the ultrasonic evaporator. It is found that the temperature difference between evaporation and heating steam is the dominant factor affecting thermal performance in this case study. Ultrasound has complicated interactions and non-linear effects in the ultrasonic evaporator. Seven machine learning algorithms are created to compare predictive thermal performance of this evaporator, including linear model, Lasso (least absolute shrinkage and selection operator), MARS (multivariate adaptive regression splines), NNet (averaging neural network), CB (Cubist model), GP (Gaussian process), and SVM (support vector machine). The SVM and NNet models among these seven models can provide accurate prediction of overall heat transfer coefficients in the ultrasonic evaporator.

Maximum relevance minimum common redundancy feature selection for nonlinear data

In recent years, feature selection based on relevance redundancy trade-off criteria has become a very promising and popular approach in the field of machine learning. However, the existing algorithmic frameworks of mutual information feature selection have certain limitations for the common feature selection problems in practice. To overcome these limitations, the idea of a new framework is developed by introducing a novel maximum relevance and minimum common redundancy criterion and a minimax nonlinear optimization approach. In particular, a novel mutual information feature selection method based on the normalization of the maximum relevance and minimum common redundancy (N-MRMCR-MI) is presented, which produces a normalized value in the range [0, 1] and results in a regression problem. We perform extensive experimental comparisons over numerous state-of-art algorithms using different forecasts (Bayesian Additive Regression tree, treed Gaussian process, k-NN, and SVM) and different data sets (two simulated and five real datasets). The results show that the proposed algorithm outperforms the others in terms of feature selection and forecasting accuracy.

On switching response surface models, with applications to the structural health monitoring of bridges

Structural Health Monitoring (SHM) is the engineering discipline of diagnosing damage and estimating safe remaining life for structures and systems. Often, SHM is accomplished by detecting changes in measured quantities from the structure of interest; if there are no competing explanations for the changes, one infers that they are the result of damage. If the structure of interest is subject to changes in its environmental or operational conditions, one must understand the effects of these changes in order that one does not falsely claim that damage has occurred when changes in measured quantities are observed. This problem – the problem of confounding influences – is particularly pressing for civil infrastructure where the given structure is usually openly exposed to the weather and may be subject to strongly varying operational conditions. One approach to understanding confounding influences is to construct a data-based response surface model that can represent measurement variations as a function of environmental and operational variables. The models can then be used to remove environmental and operational variations so that change detection algorithms signal the occurrence of damage alone. The current paper is concerned with such response surface models in the case of SHM of bridges. In particular, classes of response surface models that can switch discontinuously between regimes are discussed. Recently, it has been shown that Gaussian Process (GP) models are an effective means of developing response surface or surrogate models. However, the GP approach runs into difficulties if changes in the latent variables cause the structure of interest to abruptly switch between regimes. A good example here, which is well known in the SHM literature, is given by the Z24 Bridge in Switzerland which completely changed its dynamical behaviour when it cooled below zero degrees Celsius as the asphalt of the deck stiffened. The solution proposed here is to adopt the recently-proposed Treed Gaussian Process (TGP) model as an alternative. The approach is illustrated here on the Z24 bridge and also on data from the Tamar Bridge in the UK which shows marked switching behaviour in certain of its dynamical characteristics when its ambient wind conditions change. It is shown that treed GPs provide an effective approach to response surface modelling and that in the Tamar case, a linear model is in fact sufficient to solve the problem.

Timing Foreign Exchange Markets

To improve short-horizon exchange rate forecasts, we employ foreign exchange market risk factors as fundamentals, and Bayesian treed Gaussian process (BTGP) models to handle non-linear, time-varying relationships between these fundamentals and exchange rates. Forecasts from the BTGP model conditional on the carry and dollar factors dominate random walk forecasts on accuracy and economic criteria in the Meese-Rogoff setting. Superior market timing ability for large moves, more than directional accuracy, drives the BTGP’s success. We explain how, through a model averaging Monte Carlo scheme, the BTGP is able to simultaneously exploit smoothness and rough breaks in between-variable dynamics. Either feature in isolation is unable to consistently outperform benchmarks throughout the full span of time in our forecasting exercises. Trading strategies based on ex ante BTGP forecasts deliver the highest out-of-sample risk-adjusted returns for the median currency, as well as for both predictable, traded risk factors.