Creating a Performance Model

We may regard the present state of the universe as the effect of its past and the cause of its future. An intellect which at a certain moment would know all forces that set nature in motion, and all positions of all items of which nature is composed, if this intellect were also vast enough to submit these data to analysis, it would embrace in a single formula the movements of the greatest bodies of the universe and those of the tiniest atom; for such an intellect nothing would be uncertain and the future just like the past would be present before its eyes.

— Pierre Simon Laplace, A Philosophical Essay on Probabilities

As humans we've evolved a certain degree of credulity (a tendency to be too ready to believe that something is real or true). Our natural willingness to believe in something without evidence has distinct advantages to our survival. The information ingrained in culture was the original method of storing and transmitting vital information for our survival. Using songs, stories and art, our elders and cultural leaders passed along the information we needed to learn from past mistakes. If a member of your tribe ate some poisonous berries and became terribly ill, this information would ingrain itself to a cultural belief. After a generation, the evidence of this interaction has been long gone, but the belief would remain, protecting tribe members from being poisoned. We're particularly credulous to negative and hazardous information for obvious reasons. The risk to reward for not believing a hazard exists, whether it does or not, is incredibly skewed. Credulity has an equilibrium of course, as too much of it can lead to exploitation and false beliefs, but the fact remains that belief was essential to our survival as a species.

"Culture enables the social transmission of information, and our willingness to believe this information without scrutiny is biological"

Tribes represented hubs of information, and some hubs would communicate, but our ability to network common beliefs was limited to our immediate geography. To date, we still learn and transmit information culturally, but our networks have expanded. Story telling has turned in to blogs, podcasts and various forms of social media. Thanks to the internet, we are no longer bound by geography. Our ability to share beliefs has increased tremendously, but we are still susceptible to the same pitfalls of credulity.

As coaches, much of what we do is passed on through tradition, and some more than others feel the need to objectively confirm commonly held beliefs. We're likely to hold beliefs without evidence to confirm them, especially if something negative happens. Our subjective ability to evaluate the outcome and make a rational judgement is therefore somewhat compromised instinctively. We feel we know why a certain training program worked, or that a certain intervention is what caused an improvement in performance, but we've evolved to put the pieces to our beliefs together and not fully understand it. The mental stress caused by the contradiction of a belief forces humans to refute, reject, and seek moral support from other believers.

"The image of the world around us, which we carry in our head, is just a model. Nobody in his head imagines all the world, government or country. He has only selected concepts, and relationships between them, and uses those to represent the real system" - Forrester, 1971

Your beliefs are your model

A coach may not see the link between a model and their beliefs, but in fact, these two items are the same. A coach builds their program by organizing a sequence of stimuli to achieve an optimal result. The sequence, and relative importance of the components, are built according to the coaches inner belief model. 

If a coach believes speed, endurance, mental toughness, player skill, team dynamics, or any other key performance indicator (KPI) is crucial to performance, then they will go at lengths to train these abilities, in proportions roughly equal to their believed importance. Progress will be monitored continuously in relation to this created mental model. Presented with information that contradicts our ideologies we may refute and discredit the information and find blame in an athlete or individual. Some coaches may change their models frequently at a moments notice with little evidence, other coaches may not budge in the face of overwhelming evidence. To a certain extent we all fall somewhere on this spectrum. There is an optimal amount of credulity and flexibility we all require as coaches, but not all information was created equally, and the ability to accurately judge the quality of information that influences and changes our views is a critical skill. A coach who requires a very high standard of evidence(overly academic) may lose out on an emergent trend that could give his/her athletes an advantage. In contrast to that, a coach with no standards for evidence may lead their athletes on a wild goose chase, wasting time believing methods that have no or even detrimental effects and even worse that coach may refute/discredit the very evidence they would require to change their minds. The sweet spot here is the essence of being a data driven coach. You generate your own data and form conclusions to corroborate or refute commonly held beliefs, create models, share data with other coaches, challenge your biases, and test new paradigms when the data directs you to. While the process of being data driven is founded in scientific principles, you easily make concessions for practical purposes to preserve the structure of quality workout sessions. Your coaching practice becomes a hub for synthesizing great new insights. The key to quality coaching are robust models grounded in quality data.

The process of building an inner coach model may look like this:

As humans we build mental models to represent our external realities. These models play a major role in reasoning, decision making, and how we anticipate future events. We all possess models for coaching, a type of relational matrix we use to look up how we should expect things in our world to interact. The first mental models we built towards the sport we coach most likely came from our initial interactions with the people that coached us.

First time coach - Starting point - What I did as an athlete.

KPI 1 - 30% 

KPI 2 - 25%

KPI 3- 20%

KPI 4 - 10%

KPI 5 - 15% 

You may not explicitly express this model, but you follow it intuitively because that is the way you were coached and your culture has been ingrained. Your experiences as an athlete lead you to believe that these KPI's worked. So in essence you matched a model with the memory of a real outcome. 

As you gain experience, converse with peers, expose yourself to new ideas, your model evolves. You've essentially expanded your data set, added new variables, modified beliefs, and have matched this with new outcome variables in order to validate your new ideas. 

After 10 years of coaching, your model now looks like this:

KPI 1 - 5%

KPI 2- 10%

KPI 3 - 10%

KPI 4 - 30%

KPI 5 - 10%

KPI 6 - 15%

KPI 7 - 20%

Modulating factors: 

Sleep, stress, social, nutrition

Additional Injury Risk Factors:

Actue:Chronic training load, posture, mechanics

The model has grown in size and complexity! You've discovered other factors you had never considered before. They are more specific and honed in, while the general factors remain to an extent. You've noticed that there are interacting variables as well, that may limit progression of performance. This is somewhat of a simplification of how we each form our methods and models. At its core it reflects a simple process of aggregating information (data) and the matching of an outcome.  

Modelling your beliefs
Credulity, cognitive dissonance and cultural bias, all reduce our ability to objectively assess information as coaches. In my previous blog post, I wrote about the coach of the future being driven by algorithms. These algorithms (or models) would diminish our dependence on belief and enhance our ability to understand performance. Computer models can greatly enhance our ability to form conclusions based on our observations and help to minimize the bias we may have accumulated over the years.  A simple machine learning algorithm can explore 100 dimensional space with ease, and discover patterns and connections that we have never explored. But how do we start to formalize this process and become data driven?

By now I hope that I've convinced you that our brains have evolved to help our survival as a species, but not necessarily to have the best coaching minds! Being a data driven coach gives you access to the best of both worlds. We can enhance our productivity and accelerate the rate at which we learn and grow as a culture. Humans have a great advantage when it comes to cognition. The ability to take in an immense amount of data in the form of many senses and distill this information in to mental models. Computers are limited in scope when it comes to data representations and interpretations from our environments. Luckily it's not a competition between both, just as the chess masters have figured out that their best competitions are with computers at their sides.

So you want to be a data driven coach?
That's great! Before getting started there are a few things I'd recommend:


1. Be open minded
The best coaches in the world are constantly looking to challenge training convention. We should be open minded to any possible training paradigm, so long as the logic is sound, and the data is of a high quality . My biggest pet peeve is when an athlete or coach presents a concept as fact with little to no evidence other than "well this athlete is really good and they did this". So what!And based on what?! Be prepared to change your mind when presented with the right argument and data. Be open to your bias, we all have them, it comes with the human territory.


2. Map out your data needs
When thinking about collecting data to create a model, we need to gather a wide range of information. If we want to check our own biases, we'll need to collect data on the elements we may suspect contribute minimally, if only to rule out our own biases. Think about what may contribute to your performance. Think of the entire spectrum and write these down. These are your KPI's. These variables should explain nearly all the differences you see in performance, and if they don't you may be missing a factor.


3. Measure performance
Use a measure that is easy to collect, minimally disruptive to training and something that you plan on reproducing over the course of a season. Talk to other coaches, and ask what they measure and why. Make sure to reduce the possible sources of error that could be introduced to your data set by careless data collection. Nothing ruins a dataset faster than poor measurement practices. Be strict with your standards. You'll also need a measure of absolute performance that you are trying to model. Ideally this would be a competition performance so that part is already done for you.


Once you've collected your data, the real fun starts. Here is an example of a model I created for the 300m sprint, using a group of collegiate athletes that I coach. I theorized that the 300m could be predicted by measuring only 4 sub components. The data was collected over a single week of training, and competition results from an early meet were used to create the model.

Max V : The maximum sprinting velocity measured through a flying 25m.

Lactate Power: A repeated amount of work done at high intensity with incomplete recoveries. It could also be thought of as Intensive Tempo.

Lactate Tolerance: The ability to tolerate a high level of lactate. An extended sprint, which could be referred to as Special Endurance.

200m Time Trial: This is my marker of speed endurance.

In the graph below we have several regression variations and the breakdown of factor importance. I used 4 different mathematical methods of creating these models, and you can see that each presents the importance of each factor in a slightly different way. I'll talk about the benefits of each method in a future post.

Using these models I was able to predict the performances of a competition within 0.2s. It also lead me to change my training paradigm mid season, and it produced some truly outstanding results. The model is far from perfect, but it still contributed significantly to a successful season. Although this model was simple, and used to adjust training emphasis, the complexity of algorithms related to performance will continue to grow. 

In my next blog post, I'll go through some of the specifics about creating the model I just showed you and how I specifically used the information to make more informed decisions. But in the meantime, start collecting some data!

The future of coaching - how algorithms will enhance performance

Welcome to my first blog post. I decided to start this blog to get some ideas and concepts I've been working on out in to the world, get some feedback, engage in some discussions, and hopefully spark some new ideas. This blog will cover many topics related to data science and how it can apply to performance and coaching. I will present and discuss much of the data that I have collected over the years, but more importantly will discuss its implications to improving performance. I will make available all data sets I will discuss, and any analysis programs that I have written. I write programs in Python for those of you who want to follow along. I hope that some of these concepts, ideas, and models can be of use to any one looking to start making data driven decisions.

Before I dive in to the data (all I really care about!), I'll tell you a bit about my own experience. For the past 5 years I've been an assistant track coach at Dalhousie University working with the sprinters, and during the past 6 years, I've worked as a performance analyst and biomechanist at the Canadian Sport Centre Atlantic. In 2012 I was a part of the mission staff for the London Paralympic Games with Team Canada, and in 2016 I was team staff with Canoe/Kayak Canada for the Olympic games in Rio.

My job requires me to gather, store, sort, scrape, filter, wrangle, analyze, and present all sorts of performance metrics in an effort to improve performance. When it comes to data analysis I've seen the good, the bad and the ugly.

I'd like to kick things off by exploring what coaching might look like in the future, and specifically how this relates to data science.

Don't fear the machines

We live in a very interesting time with respect to how technology is being integrated in to our lives. Machines are starting to become more intelligent (if you can even call it that), and more and more of our tedious tasks are being replaced by algorithms. There is a growing fear that advances in A.I. will not only take over all of human productive employment but also the world! Although this topic has gained a lot of attention lately I haven't heard mention of how A.I. may be applied to coaching. I can hear you now, "A computer could never replace me!". Well don't you worry, I'm here to tell you that the future of coaching is not only safe,it is so much more awesome than you think.


To understand how this utopian future could look, I'd like to use an example from the first sport to be dominated by machines. Chess.

In May of 1997 Deep Blue became the first computer system to defeat a world champion, an amazing accomplishment by the team at IBM and the supercomputer.

The power of Deep Blue came from its ability to analyze every possible outcome from any move being played, similar to what a chess master would do but much better. Deep Blue was also programmed with the insights of many previous chess masters, in a way contained the knowledge of many champions. So now that computers reign supreme, how has the game of chess changed? Heres the amazing thing, the best chess players are not computers. The best chess players are humans using computers (called advanced chess). The ability to choose the outcome based on heightened feedback really resonates with me. I believe this outcome gives us insight in to the future of coaching.

What does the coach of the future look like?

I'd like to propose to you a vision of what a coach looks like in the future.

The coach of the future will.....

Choose the algorithm that suits his personal style and experience
The development of really robust algorithms will require some amazing data sets. These data sets do not exist today. At best, each coach, club, or organization collects a mish mash of data, whether is performance data, sleep, stresss, nutrition, heart rates, game statistics. Although the quality of the data is quite high in some instances, the data is not collected with the purpose of creating robust models. Companies will specialize in the development of performance models as a service. No model will be 100% accurate, and different models will offer different advantages based on the biases the data sets yield. A coach will become an expert in choosing a model that fits with his personal philosophies and that the user experience fits his level of expertise.


Do you want to minimize injury? Then you might choose an algorithm based on data metrics that come in from 4 separate devices and 1 assessment screen.


Understand that the algorithms and data do not run the show
At the heart of this revolution a great coach will understand that the choices and management of this system are under his control. A poorly managed system of devices and algorithms can potentially have a greater negative effect than not using any at all. The ability to choose the most efficient systems for their sport and their team will be an important coaching quality.

In the end, our tools will become increasingly complex, but will allow for more time and effort to be placed in achieving even greater feats.


Ultimately understand their sport more than any coach before
Advances in these technologies will allow coaches to reach a higher level of understanding of the key metrics which drive a team's or athletes performance. Our ability to communicate and share insights will also be enhanced. The community of coaches will learn at an unprecedented level.

See the big picture
At first algorithms will be fragmented. You'll have your injury prevention metrics, specific and unspecific performance variables, psycho-social variables, ect... All in separate products, but gradually this will evolve in to a unified theory. Coaches of the future will understand how these systems fit in to the bigger picture of performance.

Learn from an Online Education platform
Our education system is outdated. Online courses are starting to rival university level education in quality at a fraction of the cost. This trend will continue, and allow access to resources that other coaches never could. A decentralized and low cost education will increase the average level of competence in coaching.


You see, coaching in the future resembles what coaching is today. You have your distinct philosophies in various areas that come together to inform your decision making on a daily basis. These personal models are built upon your experience as a coach, athlete, and from mentors and peers. Your network may span 50 people across your career, involve 10 000 hours of real life experience, and have trialed and erred for many many seasons. The power of the machine is that it can aggregate experience and link networks at an exponentially faster rate. As of October 2016, Google's self driving car had accumulated over 300 years of driving experience, with the ability to share any new insights it gains. Having access to these catalogs of data will only serve to enhance a great coaches skills, increase the pace of learning in those individuals who are thirsty for new knowledge. I for one am excited for what is to come for the coaching profession.

Well I hope this peek in to the future has intrigued you. We are far from this reality but progress is being made and there are some people and organizations doing some really amazing things.

Stay tuned for my next blog post where I will go over in detail my attempt at building a training model for the 300m sprint, leverage genetic insights, and walk you through the season I used this model to drive decision making.