Tennis Australia getting on board with coaching technology

In observing professional tennis from afar I wouldn’t have thought that the use of technology was a prominent aspect of coaching in that sport. This article suggests Tennis Australia is looking to change all that. Good stuff.

New Coach Education Website

For coaches of team sports, and in particular Australian Rules Football, thought there might be some interest in a website we are just getting started called teamsportcoaching.com. We are endeavoring to make it one of the best  coach education resource sites out there and, as mentioned, with a strong focus on Australian Rules Football in the first 12 months. In a couple of months we will be launching a 6 month membership package which will give access to premium content delivered via a range of multimedia.

We would love you to drop by and check it out - click here to do so.

Technology examples in Gridiron

Here is a link to an article about various technology implementations in US Football. Some interesting philosophical challenges mentioned late in this article in terms of the introduction of new technology.

I am hoping to get a post out soon on this theme, along the lines of “change management” challenges in the coaching domain.

How a Grocery Store can help with your Sportscode Analysis – Part 2

Read part 1 of this post here.

For this you will require Sportscode and the Magnum Opus demo software (Windows only) that can be downloaded here

In Sportscode, combine game timelines or make a database that contains all the instances of possessions from the back third. Make sure this is the only row in the timeline. Go to File > Export > Edit List and save to your desktop. Open the file in excel and remove all the header rows and all columns except those that contain the actual labels. Save the file as a tab delimited.

Now the more complex part.

In Windows open up the Magnum Opus software. Click on “import data” and in the “file of type” filter enable all files. Navigate to your saved excel tab delimited file and click open. For the import data format option choose “Item List File” and then just click “Import Now”.

The screenshot above should now be displayed. In it you will see all your labels from Sportscode listed in both left and right hand columns. Other data you can see include how many “cases” you have imported which should correspond to the number of instances in your original sportscode timeline (in the screenshot it is 59). There are various filters such as “search for”, “search by” and “filter out”. I normally change the “search by” filter to “strength”, the meaning of which I will explain in a moment. I also change the “filter out” setting to “None”.

Ok, now we move on to selecting the labels that we want to run in our analysis. The column titled “values allowed on RHS” are what we termed in the previous post as “outcome” variables. In the screenshot above I have selected the “SOG” label (Shot On Goal) as my outcome variable. I have then selected all the labels in the column with the title “values allowed on LHS”. By using this combination I want to find out which of my labels on the left hand side column have the strongest association with the label on the right hand column – SOG. In real world language I am looking for events during the possession chain that maximize the chance of a shot on goal resulting from that chain.

When these settings are all in place click the green “Go” icon. Select a location to save the output text file. The text file should then automatically appear and look something like the screenshot below.

Lets go through some of the data. It has found 100 rules which was the limit I had set before running the output and it has ordered the rules by “Strength”. The rules are expressed in the format “Label 1 & Label 2 -> SOG”. This is saying that the rule is looking at cases where both label 1 and label 2 occurred and a shot on goal is the outcome variable. The data below this rule is the critical information. Here are the meanings of some of the terms you see there:

Coverage: the number of times in the data set that the combination of labels on the left hand side of the equation has occurred. This is expressed as both a raw number and as a decimal percentage.

Support: The number of times in the data set that the combination of labels on the left hand side and the right hand side has occurred. This is expressed as both a raw number and as a decimal percentage.

Strength: “Support” divided by “Coverage”. This is the important number because this figure tells us the likelihood that when a certain combination of labels occur that a SOG will also occur. In the screenshot this figure is 1.000 which means that every time those combination of labels appeared so did a SOG.

Lift: Remember that in our data set we had 59 cases. 32 of those cases had the SOG label in them which equates to 0.54 of cases. The “Lift” figure is telling us how many times above the 0.54 figure the stated rule “lifts” the SOG probability. So for a strength of 1.000 this is 1.84 times higher than the strength that would be expected if no relationships existed between the labels on the left and SOG on the right.

Leverage : Not really sure!

If this doesn’t make much sense there is a tutorial section under the help menu in the Magnum Opus software that explains in really good detail along with many other things I haven’t covered.I would recommend having a play around as well to get a feel for the different sorts of results you can derive based on certain settings.

So, if we go back to our output file, the main focus is on the “strength” value as this will indicate the combination of labels that increase the probability of a shot on goal occurring. As I mentioned in the previous post we could then create these combinations of labels in a code matrix organiser back in Sportscode to enable us to access the video footage of these instances for further insight and understanding.

There you have it. The use of data mining, in our case “market basket” or “assocation rule” discovery, integrated with Sportscode and the analysis process. If you feel this technique may be applicable to your sportscode data then have a crack at this and I let me know how you went.

How a Grocery Store can help with your Sportscode Analysis

A few years ago I remember data mining being the next big thing in sports analysis. I am not sure that has actually come to fruition yet although I do believe AC Milan has supposedly used some data mining techniques to help reduce injuries. There is also some work being done with the data that player tracking systems now produce to see if data mining can uncover patterns of player positioning – ie defensive shape in team field sports – that lead to positive or negative outcomes in a match.

When I started doing some research into data mining one technique that caught my eye was something called a “market basket analysis”. The background behind its use was to identify trends or associations in items people purchased at the grocery store. For example it may reveal that when people had bread in their “basket” then there was a greater likelihood they also would have butter. This example may seem obvious but often there are hidden associations in the items people have in their shopping basket. The benefit of this information for the grocery stores was how they might place their goods on the shelves to increase the chances of customers buying them as well as other marketing strategies such as placement of advertising signs etc.

How does this relate to Sportscode and analysing your data I hear you ask? Well consider this situation:

You are a soccer coach/analyst using Sportscode. As part of your analysis you code every time your team wins the ball in the defensive third, resulting in an instance in the timeline. For each instance you also have a multitude of labels that describe the events that occur during the possession until a shot on goal, turnover or out of play. These labels may include field position, passing types, players, crosses and the outcome of the play. This could result in 10 or more labels for each instance. Say you do this over 10 matches and have 300 instances coded with 10 or more labels. That is a fair amount of data!

Now whilst doing the coding you may have formed a subjective view on the sorts of events that lead to positive outcomes (say shot on goal) or negative (turnover) and this you can utilise to improve team performance. It may be obvious that when your team uses good width and then crosses into the box they generate a shot on goal most times. However what if there were more subtle patterns or associations in your data? You could run a matrix organiser with every possible combination of labels and operators (and/or/not) to find out but that would probably occupy the rest of your lifetime which I’m assuming is not an ideal option.

By using “market basket” analysis software there is a much quicker way to find out. Let me explain how it can work with your sportscode data. Let’s say that your label for “shot on goal” is the equivalent of the “bread” item in a shoppers basket as in the example we gave earlier. What we want to know is what other labels (ie events in the possession such as passes) increase the likelihood that a shot on goal will occur? We may find that crosses will. So we get to the statement “if we cross the ball we increase our chance of a shot on goal”. Another way of putting this in the market basket terminology would be “if there is a cross (ie bread) in the basket it increase the likelihood of a shot on goal (ie butter) also being in the basket”. By running your sportscode data through a market basket software program it will automatically tell you what the strongest associations with the outcome variable (shot on goal) are.

Does this make sense? The associations can also be combinations of labels such as when the labels “cross” and “left wing” appear then there is a higher chance of the label “shot on goal” also appearing. Stated in plain english this might read “when we cross the ball from the left we have a better chance of generating a shot on goal”. Obviously it would also be possible to find associations related to unsuccessful outcomes such as turnovers. This may reveal what your team needs to work on and do better.

Once you have found the associations then the next logical step is to make a code matrix organiser to generate these associations so that you can then analyse the video footage to gain further insight. From our examples above we could use the matrix organiser to make the combination of “Cross” AND “Left WIng” AND “Shot on Goal” and then run the matrix, bring up the footage and away we go!

Now that is the theory and general process behind this. If you are interested in more specific step by step details and the data mining software required then read my next post – hopefully within the next week!

Video Games for Player Learning

This link is to a blog post a while back on how playing sports video games can actually result in “learning transfer” to the real-world sporting environment.

I believe it is now common practice for gridiron coaches to have their players learn their playbook by playing the Madden Football video game loaded with specific plays. And current technology has enabled the creation of 3D and virtual reality environments for professional players to “immerse” themselves in.

Look for this field to really take off in the next few years as a legitimate training and learning tool in elite level team sport.

Carlton Football Club – Use of iphone’s

Video interview here .

Interesting to hear their Academy coach talk about the “technophobia” of the older generation of coaches compared with the young players who have grown up with the technology.