User contributions in the form of posts, comments, and votes are essential to the success of online communities. However, allowing user participation also invites undesirable behavior such as trolling. In this paper, we characterize antisocial behavior in three large online discussion communities by analyzing users who were banned from these communities. We find that such users tend to concentrate their efforts in a small number of threads, are more likely to post irrelevantly, and are more successful at garnering responses from other users. Studying the evolution of these users from the moment they join a community up to when they get banned, we find that not only do they write worse than other users over time, but they also become increasingly less tolerated by the community. Further, we discover that antisocial behavior is exacerbated when community feedback is overly harsh. Our analysis also reveals distinct groups of users with different levels of antisocial behavior that can change over time. We use these insights to identify antisocial users early on, a task of high practical importance to community maintainers.

Most of Markov Chain Monte Carlo (MCMC) and sequential Monte Carlo (SMC) algorithms in existing probabilistic programming systems suboptimally use only model priors as proposal distributions. In this work, we describe an approach for training a discriminative model, namely a neural network, in order to approximate the optimal proposal by using posterior estimates from previous runs of inference. We show an example that incorporates a data-driven proposal for use in a non-parametric model in the Anglican probabilistic programming system. Our results show that data-driven proposals can significantly improve inference performance so that considerably fewer particles are necessary to perform a good posterior estimation.

Traditional graph-based semi-supervised learning (SSL) approaches, even though widely applied, are not suited for massive data and large label scenarios since they scale linearly with the number of edges $|E|$ and distinct labels $m$. To deal with the large label size problem, recent works propose sketch-based methods to approximate the distribution on labels per node thereby achieving a space reduction from $O(m)$ to $O(\log m)$, under certain conditions. In this paper, we present a novel streaming graph-based SSL approximation that captures the sparsity of the label distribution and ensures the algorithm propagates labels accurately, and further reduces the space complexity per node to $O(1)$. We also provide a distributed version of the algorithm that scales well to large data sizes. Experiments on real-world datasets demonstrate that the new method achieves better performance than existing state-of-the-art algorithms with significant reduction in memory footprint. We also study different graph construction mechanisms for natural language applications and propose a robust graph augmentation strategy trained using state-of-the-art unsupervised deep learning architectures that yields further significant quality gains.

Nowadays large amounts of GPS trajectory data is being continuously collected by GPS-enabled devices such as vehicles navigation systems and mobile phones. GPS trajectory data is useful for applications such as traffic management, location forecasting, and itinerary planning. Such applications often need to extract the time-stamped Sequence of Visited Locations (SVLs) of the mobile objects. The nearest neighbor query (NNQ) is the most applied method for labeling the visited locations based on the IDs of the POIs in the process of SVL generation. NNQ in some scenarios is not accurate enough. To improve the quality of the extracted SVLs, instead of using NNQ, we label the visited locations as the IDs of the POIs which geometrically intersect with the GPS observations. Intersection operator requires the accurate geometry of the points of interest which we refer to them as the Geometries of Interest (GOIs). In some application domains (e.g. movement trajectories of animals), adequate information about the POIs and their GOIs may not be available a priori, or they may not be publicly accessible and, therefore, they need to be derived from GPS trajectory data. In this paper we propose a novel method for estimating the POIs and their GOIs, which consists of three phases: (i) extracting the geometries of the stay regions; (ii) constructing the geometry of destination regions based on the extracted stay regions; and (iii) constructing the GOIs based on the geometries of the destination regions. Using the geometric similarity to known GOIs as the major evaluation criterion, the experiments we performed using long-term GPS trajectory data show that our method outperforms the existing approaches.

To reiterate, the space is coloured according to whether the machine learning technique predicts it belongs to the red or the blue class. For example, if a model predicts a high probability that a region is blue, then we shade that area darker blue). The line between coloured regions is called the decision boundary. The first thing you might look for is how many points have been misclassified by being included in an incorrectly coloured region. To perfectly solve this problem, a very complicated decision boundary is required.

I have 400 units and 10 groups, and I'm classifying the units' group membership using a discriminant function analysis or linear discriminant analysis. During cross-validation, I want to test that my solution is doing a better job at classifying them than chance (10%). I can get an error rate, but don't know how to statistically compare. With the hold-out approach, I can test it using Press' Q statistic or Maximum Chance Criterion. But with k-fold I don't think I can use this approach.

This weekend's race is as Dover International Speedway, a 1-mile concrete oval with steep banking in the corners. I'll give you picks and projections to help you set your DraftKings NASCAR Dover lineups. The race is scheduled for 400 laps, so finding the dominators is of utmost importance when setting your lineups this week. For more on strategy, listen to this week's NASCAR episode of On the Daily DFS and check out my Dover preview article. I'll continue my logistic regression model to predict the probability that a driver ends up with a top six and a top 10 score for this weekend's DraftKings NASCAR Dover slate. It was highly successful last weekend at Kansas, so let's see how it fares this week.

Of course, if you are an experienced Python programmer you will be able to skip this step. Even if so, I suggest keeping the very readable Python documentation handy. KDnuggets' own Zachary Lipton has pointed out that there is a lot of variation in what people consider a "data scientist." This actually is a reflection of the field of machine learning, since much of what data scientists do involves using machine learning algorithms to varying degrees. Is itnecessary to intimately understand kernel methods in order to efficiently create and gain insight from a support vector machine model?

Regression is still one of the most widely used predictive methods. If you are unfamiliar with Linear Regression, check out my: Linear Regression using Excel lesson. It will explain the more of the math behind what we are doing here. This lesson is focused more on how to code it in Python. What we have is a data set representing years worked at a company and salary.

This work focuses on dynamic regret of online convex optimization that compares the performance of online learning to a clairvoyant who knows the sequence of loss functions in advance and hence selects the minimizer of the loss function at each step. By assuming that the clairvoyant moves slowly (i.e., the minimizers change slowly), we present several improved variation-based upper bounds of the dynamic regret under the true and noisy gradient feedback, which are {\it optimal} in light of the presented lower bounds. The key to our analysis is to explore a regularity metric that measures the temporal changes in the clairvoyant's minimizers, to which we refer as {\it path variation}. Firstly, we present a general lower bound in terms of the path variation, and then show that under full information or gradient feedback we are able to achieve an optimal dynamic regret. Secondly, we present a lower bound with noisy gradient feedback and then show that we can achieve optimal dynamic regrets under a stochastic gradient feedback and two-point bandit feedback. Moreover, for a sequence of smooth loss functions that admit a small variation in the gradients, our dynamic regret under the two-point bandit feedback matches what is achieved with full information.