Search

An important technique to explore a black-box machine learning (ML) model is called SHAP (SHapley Additive exPlanation). SHAP values decompose predictions into contributions of the features in a fair way. We will show that for a boosted trees model with some or all features being additively modeled, the SHAP dependence plot of such a feature corresponds to its partial dependence plot up to a vertical shift. We illustrate the result with XGBoost., Comment: 15 pages, 5 figures

One of the main tasks of actuaries and data scientists is to build good predictive models for certain phenomena such as the claim size or the number of claims in insurance. These models ideally exploit given feature information to enhance the accuracy of prediction. This user guide revisits and clarifies statistical techniques to assess the calibration or adequacy of a model on the one hand, and to compare and rank different models on the other hand. In doing so, it emphasises the importance of specifying the prediction target functional at hand a priori (e.g. the mean or a quantile) and of choosing the scoring function in model comparison in line with this target functional. Guidance for the practical choice of the scoring function is provided. Striving to bridge the gap between science and daily practice in application, it focuses mainly on the pedagogical presentation of existing results and of best practice. The results are accompanied and illustrated by two real data case studies on workers' compensation and customer churn., Comment: 70 pages, 22 figures

Many proteins undergo conformational changes during their activity. A full understanding of the function of these proteins can only be obtained if different conformations and transitions between them can be monitored in aqueous solution, with adequate temporal resolution and, ideally, on a single-molecule level. Interrogating conformational dynamics of single proteins remains, however, exquisitely challenging and typically requires site-directed chemical modification combined with rigorous minimization of possible artifacts. These obstacles limit the number of single-protein investigations. The work presented here introduces an approach that traps single unmodified proteins from solution in a plasmonic hotspot and makes it possible to assign changes in refractive index to changes in protein conformation while monitoring these changes for minutes to hours with a temporal resolution at least as fast as 40 microseconds. The resulting single molecule data reveals that adenylate kinase employs a hidden enzymatic sub-cycle during catalysis, that citrate synthase populates a previously unknown intermediate conformation, which is more important for its enzymatic activity than its well-known open conformation, that hemoglobin transitions in several steps from its deoxygenated and rigid T state to its oxygenated and flexible R state, and that apo-calmodulin thermally unfolds and refolds in steps that correspond to conformational changes of individual protein domains., Comment: 20 pages, 4 figures

Strongly electric fish use gradients of ions within their bodies to generate stunning external electrical discharges; the most powerful of these organisms, the Atlantic torpedo ray, can produce pulses of over 1 kW from its electric organs. Despite extensive study of this phenomenon in nature, the development of artificial power generation schemes based on ion gradients for portable, wearable, or implantable human use has remained out of reach. Previously, inspired by the electric eel, we developed an artificial electric organ that generated electricity from ion gradients within stacked hydrogels and, like the eel, was optimized to deliver large voltages that exceeded 100 V. Due to its high internal resistance, the current of this power source was, however, too low to power standard electronics. Here we introduce an artificial electric organ that takes inspiration from the unique morphologies of torpedo rays for maximal current output. This power source uses a hybrid material of hydrogel-infused paper to create, organize, and reconfigure stacks of thin, arbitrarily large gel films both in series and in parallel. The resulting increase in electrical power by almost two orders of magnitude compared to the original eel-inspired design makes it possible to power electronic devices and establishes that biology's mechanism of generating significant electrical power can now be realized from benign and soft materials in a portable size., Comment: 34 pages, 10 figures, 1 table

The ctools open-source software package was developed for the scientific analysis of astronomical data from Imaging Air Cherenkov Telescopes (IACTs), such as H.E.S.S., VERITAS, MAGIC, and the future Cherenkov Telescope Array (CTA). To date, the software has been mainly tested using simulated CTA data; however, upon the public release of a small set of H.E.S.S. observations of the Crab nebula, MSH 15-52, RX J1713.7-3946, and PKS 2155-304 validation using real data is now possible. We analysed the data of the H.E.S.S. public data release using ctools version 1.6 and compared our results to those published by the H.E.S.S. Collaboration for the respective sources. We developed a parametric background model that satisfactorily describes the expected background rate as a function of reconstructed energy and direction for each observation. We used that model, and tested all analysis methods that are supported by ctools, including novel unbinned and joint or stacked binned analyses of the measured event energies and reconstructed directions, and classical On-Off analysis methods that are comparable to those used by the H.E.S.S. Collaboration. For all analysis methods, we found a good agreement between the ctools results and the H.E.S.S. Collaboration publications considering that they are not always directly comparable due to differences in the datatsets and event processing software. We also performed a joint analysis of H.E.S.S. and Fermi-LAT data of the Crab nebula, illustrating the multi-wavelength capacity of ctools. The joint Crab nebula spectrum is compatible with published literature values within the systematic uncertainties. We conclude that the ctools software is mature for the analysis of data from existing IACTs, as well as from the upcoming CTA., Comment: 34 pages, 47 figures