Machine Learning • machine Learning in betting and in football

Machine Learning - machine Learning applied to bets in football

Machine Learning, or in our good Portuguese: machine learning, is what enables the bookmakers set the odds – or quotes – that you see always when you open a site of any of them. Maybe you have the impression that this is a thing of the asian Silicon Valley, but apply the learning machines, in betting and in football is increasingly more accessible, and perhaps this article is the watershed for a fantastic learning experience that you ever imagined to have.

And what is this Machine Learning or machine Learning?

What is Machine Learning

If you search on Wikipedia for some explanation more formal we will have something more or less like this:

It is the ability of computers to learn and make decisions without being exactly programmed to do so. You learn through the examples, pondering mistakes and successes through mathematical algorithms.

See, I sapequei a bold in the “learn by example”, because it is precisely through the amount of examples, or the amount of samples that we offer to the learning algorithm of the machine that he’ll be able to actually learn something.

An example of an idiot of machine learning

Example of algorithm

Nothing better than an example, those as well had the same, for this to be made very clear. Let’s say that I want to make a prediction of classification and, therefore, I want to predict whether a thing can be:

  • Our friendly designer, the Markin;
  • A bottle of beer;
  • Or a cow.

To make this forecast I need to bring hundreds or even thousands of examples of Markinhos, bottles of beer and cows. And the more relevant features I be able to bring in my examples, the better will be my model of machine learning.

The variables in machine learning: no. of paws, muge?

Let’s say that I, with all my incompetence, only able to bring two variables:

  • The amount of the legs;
  • Muge?

Therefore, we have there a numeric variable discrete what is the amount of paws, and a binary variable which has this name because it only assumes two values: 0 for no and 1 for yes. You see, as would be our data set that we are using to train our model:

Wonderful! After you have shown for our algorithm a caralhada of Markinhos, Beer Bottles, and of Cows, the model will be able to find a pattern through the variables and hence, to see a “new thing” like this:

He will say: “Info! Saporra here is a cow! With 99.99% certainty!” And see that to reach such a decision, was used an algorithm widely used in machine learning: a decision tree.

The algorithms are the tools for the solution of problems

The decision tree is an algorithm because it is a set of rules and logical operations and mathematical that allows us to solve a particular problem. In other words, algorithms are like tools, and you need to seek the best tool for a given problem. A hammer can be great for pounding a nail, but a shit for straining a coffee break.

Thus, the decision tree has been asking questions for each of the variables and, depending on the answers, a different classification is given to this new thing that until then had not rating any.

Machine learning in football betting

Machine learning and bets on football

This is the time that you should ask yourself:

Okay! But what cocks the Markinho, or the beer or the cow has to do with Machine Learning in the betting on soccer?

And the answer is simple: in our previous issue, our challenge was to classify new things into three categories, which were the Markinho, the bottle of beer and a cow. Now, as bettors, our challenge may be to sort a match that will happen between:

  • Principal;
  • Tie;
  • Visitor.

Or, even if the match will have more or fewer than 2.5 goals, or even if a particular Handicap will hit. And more importantly: to calculate the probabilities of each of these classifications, because once we have the probabilities, we can convert it into odds, and then convert it to odds we know if a bet or not it has positive expected value.

The variables in machine learning in football

The variables in machine learning

There is a field of study in data science called ‘Feature Engineering’, the translation to Portuguese is pretty bad: engineering resources. Thus, to understand Feature Engineering as their ability to acquire and develop new variables for your model of machine learning.

Thus, to create a model to predict the outcome of a football match have variables such as:

  • No. of paws
  • Muge?

Has no value! Because these variables do not help us with anything in our new issue. In football, do a predictive analysis requires more variables and also requires a greater complexity to obtain them.

At the beginning of 2019, we here Club we did a selection process to hire a new scientist data.See only as was the data set used to train the model of machine learning that we use as a challenge in the selection process. I’m going to bring a sample of the first 5 lines:

An example of a set of data used for machine learning in football

This set consisted of 30 variables, which are they:

  • ‘home_name’: Name of the principal,
  • ‘away_name’: Name of the visitor,
  • ‘home_score’: Goals made by the principal in the match,
  • ‘away_score’: Goals made by the visitor at start-up,
  • ‘final_result’: This is the variable that we want to predict, this is the end result, being H (Home) Win of the Principal, D (Draw) Tie, and, finally, (Away) visitor,
  • ‘time’: Time in unix format,
  • ‘home_pos’: The position of the principal prior to departure,
  • ‘away_pos’: the position of The visitor prior to departure,
  • ’round’: the round of The tournament,
  • ‘home_last5all_home’: goals of the principal in the last 5 matches,
  • ‘home_last5all_home_win’: No. of wins at home in last 5 matches,
  • ‘home_last5all_home_draw’: No. of draws of the principal in the last 5 matches,
  • ‘home_last5all_home_lose’: No losses of principal in the last 5 matches,
  • ‘away_last5all_away’: the Balance of the goals of the visitor in the last 5 matches,
  • ‘away_last5all_away_win’: No. of wins of visitor in the last 5 matches,
  • ‘away_last5all_away_draw’: No. of draws the visitor in the last 5 matches,
  • ‘away_last5all_away_lose’: No. of defeats of the visitor in the last 5 matches,
  • ‘last5all_home_away_dif’: The difference of the balance between the teams, ie: ‘home_last5all_home’ – ‘away_last5all_away’
  • ‘fifa_home_ova’: Score General Principal in Fifa
  • ‘fifa_home_att’: Score attack of the Home team in the Fifa
  • ‘fifa_home_mid’: Score from the middle of the field for the Client’s in Fifa
  • ‘fifa_home_def’: Score defence of the Principal in Fifa
  • ‘fifa_away_ova’: Score General Visitor in Fifa
  • ‘fifa_away_att’: Score attack Visitor in Fifa
  • ‘fifa_away_mid’: Score of the middle field of the Visitor in Fifa
  • ‘fifa_away_def’: Score defence of the Visitor in Fifa
  • ‘elo_home_score’: Score Link for the Client
  • ‘elo_away_score’: Score Link from the Visitor
  • ‘tfm_value_home’: market Value of the cast at home in Euros
  • ‘tfm_value_away’: market Value of of the cast as visitor in Euros

Your ability and creativity are able to create and build good variables will be critical to the success of your model of machine learning in sports betting.

An interesting example was one of the posts of the Users of the Pinnacle, where they said that some decades ago some players started to have a competitive advantage over homes because there were inserted the variable ‘weather conditions’ to your model. However, soon the houses if you have upgraded, by entering this variable also and eliminating this advantage which had been gained.

Why you should study Machine Learning as a player?

Why study machine learning

You probably already have noticed that for adventure in the area of machine learning or data science as a whole you will need to learn a programming language. Currently, the most recommended by the community are:

  • Python
  • R

And it comes precisely from there the biggest advantage in this way: even if you can’t get good results the walk will be worth it.

Learning programming is useful for life

If you dedicate years of your life to study the trading sports, operate software like GeeksToy, to understand resistance, weight of money, time bombs, moments of the game, among others concepts of trading sports; hardly you will be able to carry this knowledge to other areas of life if you fail to become a trader profitable.

However, this problem does not occur here. Because what you will learn in Python, which is the language that we use and recommend, as well as all the knowledge of machine learning, can be applied in various area of your life, be it personal or professional.

Once, my great friend ‘J’ told me something that I agree very much: ‘programming is the new English’.

So, if before we needed the English to distinguish ourselves professionally, the same is already happening with the programming which is also a language. Are you learning a way to talk with your computer what to do.

Remember: this is not a thing of the asian Silicon Valley, programming is accessible to all, and learn it in the context of sports betting is very pleasurable.

The anguish of being out of time

The anguish of being out of time

As a reinforcement of the previous argument, it is distressing to dedicate your time studying something that you may not will bring you the returns that you want. And I repeat: even if you are not able to earn even a penny with sports betting through their models in the progress of learning will have been worth it.

After all, you have learned a skill that is regarded as essential for the century that we are going to face.

You will hardly break a bankroll

You will hardly break a bankroll

When you create a model and then automate it, you’ll just break your bankroll if you are, with the forgiveness of the word, a part of me. Or if you’ve done some shit that allowed his program – very probably for some bug – bet in addition to the percentage that you have set.

In addition, when training a model for machine learning you will divide it into two data sets:

  • Training Set: set of training;
  • Test Set: a set of tests.

So, guess what: you can simulate the earnings of your model on the test set, which is a set that has never been seen by the model, so it is something new, completely new. If you had the proper care in avoiding the Overfitting of data, or over-adjustment, this model will have performance similar in the new data that will come.

Program is to give a series of logical instructions for your computer and it will go line by line. So, see what a wonderful thing: your computer is not going to want to bet all of your bankroll just because the World will play against Fluminense shorn.

The coolness of the logic of computers is to our favor, there is emotional, there is no heart, but only the objectivity raw of what was programmed for you.

Want to learn Machine Learning applied to bets in football?

Like? So take advantage of that this year, in partnership with the bookmaker, Pinnacle, let’s do a advanced training complete in Punting, and we’ll teach you to program in Python, analyze data, assemble your sets of data to train your model, and make predictions for soccer matches.

Click on the banner below and learn more about our Course betting Punting advanced:

In addition, for those who want to delve deeper in this area, I make the following suggestions:

  • Kaggle: a Community of data scientists, challenges, forum, courses and discussions.
  • Datacamp: one of the best portals of courses on data Science that I have met;
  • Quora: is the Yahoo Answer that got it right. Discussions of the most high level are done by there.
  • Users of the Pinnacle: this is the best blog of analytical content focused on sports betting.

I’m getting by here. See you in our course!

Leave a Reply

Your email address will not be published. Required fields are marked *