Sports Analytics Weekly by kubeia.io - 44/2023

This is a FULL xG Football Trading review written by Ian Taylor who is a full-time writer and hobbyist football trader. His views are his own.

After nearly 6 years I now think you are correct. All these horse trading pre-off is just marketing gimmick to sell his software, probably with the blessing of Betfair themselves to bring in more liquidity.I tried pre-off horse trading myself and it makes no sense. There is no rule where the odds have to be, it could go anywhere. I fail to see how staring at odds moving around somehow give you insight to the future...

In the latest SBC Podcast I am joined by Nick Goff, the professional football punter, former industry trader and member of a highly successful syndicate.

Nick has an extremely interesting back story, with industry experience and management of markets for one of the UK’s leading bookmakers shaping his knowledge and expertise.

Disillusioned with the direction the industry was taking, Nick took the plunge and decided to back his judgement by becoming a full-time bettor seven years ago.

Once again, it’s time for the Major League Soccer playoffs and penalty shootout season. Changes in the playoff format to a best of three-game first round where tie games after regulation go straight to a shootout mean that a player’s penalty taking skill may be more important than ever. 

The historical trends in league gate attendances in English football can be powerfully summarised visually using timeplotsTotal league attendances peaked in 1948/49 and thereafter declined until the mid-1980sLeague attendances across the Premier League and Football League have recovered dramatically since the mid-1980s and are now at levels last experienced in the 1950sUsing average gates to allow for changes in the number of clubs and matches, the  Premiership matches in 2022/23 averaged 40,229 spectators per match, the highest average gate in the top division since the formation of the Football League in 1888 

Financial determinism in professional team sports refers to those leagues in which sporting performance is largely determined by expenditure on playing talentFinancial determinism creates the “shooting-star” phenomenon – a small group of ”stars”, big-market teams with the high wage costs and high sporting performance, and a large “tail” of smaller-market teams with lower wage costs and lower sporting performanceThere is a very high degree of financial determinism in the English Premier LeagueAchieving high sporting efficiency is critical for small-market teams with limited wage budgets seeking to avoid relegation

The most useful summary statistic for a trended variable is the average growth rateBut there are several different methods for calculating average growth rates that can often generate very different results depending on whether all the data is used or just the start and end points, and whether simple or compound growth is assumedBe careful of calculating average growth rates using only the start and end points of trended variables since this implicitly assumes that these two points are representative of the dynamic path of the trended variable and may give a very biased estimate of the underlying growth rateBest practice is to use all of the available data to estimate a loglinear trendline which allows for compound growth and avoids having to calculate an appropriate midpoint of a linear trendline to convert the estimated slope into  growth rate

If you are interested in working with images and video and would like to introduce machine learning into your computer vision applications, then OpenCV is a library that you will need to get hold of. OpenCV is a huge open source library that can interface with various programming languages, including Python, and which is extensively used by many individuals and commercial entities. In this tutorial, you will familiarise yourself with the OpenCV library and what makes it important. 

As I love to explain option trading and quantitative concepts for non-professionals (I’m already making some nice progress with my daughters, who are 12yrs and 9yrs old), I thought of putting together a blog post that explains the very basics of volatility trading. I will try to keep the quant knowledge to the bare minimum, and the only prerequisite with regards to options trading is that the reader knows what Put/Call options are. Ready? so let’s dive in…

My journey into the world of AI didn’t begin when OpenAI released a sophisticated autocomplete model; it started with a course at Cornell University — Foundations of Artificial Intelligence. I took the course because I was into trading. I had found academic articles, that promised to give traders unprecedented returns, if you understood the technical jargon. I was determined to develop an algorithm that could trade stocks for me.

..or can you? A deceptively simple question with a complex answer – come join a mathematical journey into madness and wonder, in search of answers that might just give you a new perspective on the mathematical constructs we use in our games

BackgroundI've been running an experiment to see how well ChatGPT-Generated Portfolios perform in real-time trading. In this experiment, several versions of the portfolios that are deployed are "optimized" variants of the original ChatGPT-Generated Portfolio. But what exactly is an "optimized" portfolio? This article hopes to answer the question so that we walk away understanding the benefits of genetic optimization. We'll define some terminology, explain why we should care about GAs, and describe some benefits and drawbacks to these biologically-inspired optimization algorithms

As Andrew and I wrote in our 2021 Harvard Data Science Review article, the simplistic (and unrealistic) view of EDA as not involving any substantive a priori expectations on the part of the analyst can be harmful for practical development of visualization tools. It can lead to a plethora of graphical user interface systems, both in practice and research, that prioritize serving up easy-to-parse views of the data, at the expense of surfacing variation and uncertainty or enabling the analyst to interrogate their expectations. These days we have lots of visualization recommenders for recommending the right chart type given some query, but it’s usually about getting the choice of encodings (position, size, etc.) right. 

With machine learning all over the news, it’s an exciting time to apply the ergodicity lens to the topic.

Optimization is ubiquitous. While derivative-based algorithms have been powerful tools for various problems, the absence of gradient imposes challenges on many real-world applications. In this work, we propose Optimization by PROmpting (OPRO), a simple and effective approach to leverage large language models (LLMs) as optimizers, where the optimization task is described in natural language. In each optimization step, the LLM generates new solutions from the prompt that contains previously generated solutions with their values, then the new solutions are evaluated and added to the prompt for the next optimization step. We first showcase OPRO on linear regression and traveling salesman problems, then move on to prompt optimization where the goal is to find instructions that maximize the task accuracy. With a variety of LLMs, we demonstrate that the best prompts optimized by OPRO outperform human-designed prompts by up to 8% on GSM8K, and by up to 50% on Big-Bench Hard tasks.

How is this possible?

Tabular data is one of the most commonly used types of data in machine learning. Despite recent advances in neural nets (NNs) for tabular data, there is still an active discussion on whether or not NNs generally outperform gradient-boosted decision trees (GBDTs) on tabular data, with several recent works arguing either that GBDTs consistently outperform NNs on tabular data, or vice versa. In this work, we take a step back and question the importance of this debate. To this end, we conduct the largest tabular data analysis to date, comparing 19 algorithms across 176 datasets, and we find that the 'NN vs. GBDT' debate is overemphasized: for a surprisingly high number of datasets, either the performance difference between GBDTs and NNs is negligible, or light hyperparameter tuning on a GBDT is more important than choosing between NNs and GBDTs. A remarkable exception is the recently-proposed prior-data fitted network, TabPFN: although it is effectively limited to training sets of size 3000, we find that it outperforms all other algorithms on average, even when randomly sampling 3000 training datapoints. Next, we analyze dozens of metafeatures to determine what properties of a dataset make NNs or GBDTs better-suited to perform well. For example, we find that GBDTs are much better than NNs at handling skewed or heavy-tailed feature distributions and other forms of dataset irregularities. Our insights act as a guide for practitioners to determine which techniques may work best on their dataset. Finally, with the goal of accelerating tabular data research, we release the TabZilla Benchmark Suite: a collection of the 36 'hardest' of the datasets we study. Our benchmark suite, codebase, and all raw results are available at this https URL.

Machine Learning models are great at many tasks. However, one of the biggest challenges is that these models are not calibrated. Watch the video to find out what we mean by calibration for machine learning models and why everyone care about it.

Adding confidence measures to predictive models should increase the trustworthiness, but only if the models are well-calibrated. Historically, some algorithms like logistic regression, but also neural networks, have been considered to produce well-calibrated probability estimates off-the-shelf. Other techniques, like decision trees and Naive Bayes, on the other hand, are infamous for being significantly overconfident in their probabilistic predictions. In this paper, a large experimental study is conducted to investigate how well-calibrated models produced by a number of algorithms in the scikit-learn library are out-of-the-box, but also if either the built-in calibration techniques Platt scaling and isotonic regression, or Venn-Abers, can be used to improve the calibration. The results show that of the seven algorithms evaluated, the only one obtaining well-calibrated models without the external calibration is logistic regression. All other algorithms, i.e., decision trees, adaboost, gradient boosting, kNN, naive Bayes and random forest benefit from using any of the calibration techniques. In particular, decision trees, Naive Bayes and the boosted models are substantially improved using external calibration. From a practitioner’s perspective, the obvious recommendation becomes to incorporate calibration when using probabilistic prediction. Comparing the different calibration techniques, Platt scaling and VennAbers generally outperform isotonic regression, on these rather small datasets. Finally, the unique ability of Venn-Abers to output not only well-calibrated probability estimates, but also the confidence in these estimates is demonstrated.

Machine learning classification tasks are pervasive, ranging from differentiating between cats and dogs to diagnosing severe diseases and identifying pedestrians for self-driving cars.However, these problem definitions often overshadow the true goal of classification: facilitating informed decision-making. Merely having class labels doesn’t suffice; what’s crucial are well-calibrated class probabilities.While many data scientists gauge a model’s efficacy using metrics like accuracy, precision, and recall, these can be misleading outside of basic scenarios like the cats-vs-dogs example. Regrettably, essential topics like classifier calibration often go unaddressed in foundational machine learning courses, such as Andrew Ng’s renowned machine learning course.

This is a behind the scenes look at how we scaled ChatGPT and the OpenAI APIs.Scaling teams and infrastructure is hard. It's even harder when there are not enough graphics processing units (GPUs) left in the world to serve demand. This story is also about staying nimble enough to release new capabilities and respond quickly to a rapidly changing industry.You will leave this talk with knowledge on: Why AI isn’t a magical black box How ChatGPT gets the most out of GPUsHow to scale a company at speed and respond to unexpected demands quickly

Nvidia (NVDA.O) dominates the market for artificial intelligence computing chips. Now it is coming after Intel’s longtime stronghold of personal computers.Nvidia has quietly begun designing central processing units (CPUs) that would run Microsoft’s (MSFT.O) Windows operating system and use technology from Arm Holdings(O9Ty.F), , two people familiar with the matter told Reuters.The AI chip giant's new pursuit is part of Microsoft's effort to help chip companies build Arm-based processors for Windows PCs. Microsoft's plans take aim at Apple, which has nearly doubled its market share in the three years since releasing its own Arm-based chips in-house for its Mac computers, according to preliminary third-quarter data from research firm IDC.

Ever wondered why?

In this tutorial, you will learn the three primary reasons your validation loss may be lower than your training loss when training your own custom deep neural networks.

I agree with Murtaugh (and also with Greenland and Poole 2013, who make similar points from a Bayesian perspective) that with simple inference for linear models, p-values are mathematically equivalent to confidence intervals and other data reductions, there should be no strong reason to prefer one method to another. In that sense, my problem is not with p-values but in how they are used and interpreted.

This article provides a comprehensive study of different ways to make speed benchmarks of gradient boosted decision trees algorithm. We show main problems of several straight forward ways to make benchmarks, explain, why a speed benchmarking is a challenging task and provide a set of reasonable requirements for a benchmark to be fair and useful.

Statistical modeling is a powerful tool for developing and testing theories by way of causal explanation, prediction, and description. In many disciplines there is near-exclusive use of statistical modeling for causal explanation and the assumption that models with high explanatory power are inherently of high predictive power. Conflation between explanation and prediction is common, yet the distinction must be understood for progressing scientific knowledge. While this distinction has been recognized in the philosophy of science, the statistical literature lacks a thorough discussion of the many differences that arise in the process of modeling for an explanatory versus a predictive goal. The purpose of this article is to clarify the distinction between explanatory and predictive modeling, to discuss its sources, and to reveal the practical implications of the distinction to each step in the modeling process.

Empirical policy research often focuses on causal inference. Since policy choices seem to depend on understanding the counterfactual–what happens with and without a policy–this tight link of causality and policy seems natural. While this link holds in many cases, we argue that there are also many policy applications where causal inference is not central, or even necessary.Consider two toy examples. One policy maker facing a drought must decide whether to invest in a rain dance to increase the chance of rain. Another seeing clouds must deciding whether to take an umbrella to work to avoid getting wet on the way home? Both decisions could benefit from an empirical duty of rain. But each has different requirements of the estimator. One requires causality: do rain dances cause rain? The other does not, needing only prediction: is the chance of rain high enough to merit an umbrella? We often focus on rain dance like policy problems. But there are many important policy problems umbrella-like. Not only are these prediction problems neglected, machine learning can help us solve them more effectively.

There are lots of hurdles to overcome before you can expect to turn a profit from following a tipster. First and foremost of course it’s the tipster’s skill to identify value prices. Secondly you want the tipster to advise his bets at prices from a reputable bookmaker who does not restrict their customers. Additionally you need to deduct subscription fees from potential profits and bet placement fees if you happen to place your bets automated through a bot (like I do). Something that is not so obvious and sometimes hard to estimate is slippage, which also has a hefty influence on your bottom line.Slippage is the difference in profits between betting at advised and betting at obtained prices

Hiring competent data scientists, analysts and machine learning engineers for forecasting roles is crucial, but many candidates sabotage themselves in interviews by making amateur mistakes.

Here are the top 10 time series forecasting flubs that raise serious red flags for any technically competent hiring manager. Avoid these common pitfalls to ace your next data science interview!