[PYTHON] Use a cool graph to analyze PES data!

We will automate the data analysis of PES! Part2

Hi, this is Yajun

In this article, I tried to automate the data aggregation of PES! is continued. Now that I can collect data from images, I immediately tried "analysis"-"discussion". Here are some of the implemented functions.

• This software is created with python. ■ Reference site

■ Analysis goal

• The goal of the analysis was to model the player! Modeling a player doesn't mean making it a beautiful girl or a handsome guy. (Laughing here!) From the data, the models of "when you win" and "when you lose" are visualized with graphs. From the gap between the "winning model" and the "losing model", we will clarify the challenges and the strengths of the players!

■ Flow

1. Create a graph from each parameter
2. Gap analysis at the time of victory or defeat
3. Visualize issues for players
4. Selection of the next analysis method

■ Histogram & scatter plot

[Graph]

• With the previously created Metrics Collection Software, we have collected the match results for 30 testers. It was. What a tester! That "Chama-san"! There are more than 80 types of parameters that can be acquired. Since it is meaningless to graph all parameters, I selected four as an example this time.

Possession Ratio in the center of the attack area Number of passes Number of balls captured

![](https://i.imgur.com/VAweSJ6.png)

(Example) The scatter plot of the second square from the top and the first square from the left is a scatter plot of the vertical axis "ratio in the center of the attack area" and the horizontal axis "possession".

* Supplement

• Histogram and Scatter Plot I will explain briefly for those who do not know. In the histogram of "possession" on the upper left, the peak of the bar is coming at "50% ~ 55%", isn't it? This bar shows "how many data of the corresponding possession (%) were there". "In the results of this tester's match, there were many matches with 50-55% possession!"

[Discussion]

• Let's take a look at the graph! The scatter plot in the top row of the vertical axis "possession" shows a clear tendency. If the ratio of "center of attack area" increases, "possession" also increases, so the distribution of points tends to increase. "Number of passes" and "number of balls captured" have the same tendency.

You can see the tendency in other graphs, but it seems impossible to "clarify the difference between winning and losing draws". In many cases, the expected results cannot be obtained with just two pieces of information.

• However, this time the data was for 30 games, so it's a different story if this is 100 games. Since the amount of information for one parameter increases, good results can be seen.

Next!

■ Heat map analysis

[Graph]

• We will visualize the information on ball capture and ball lost.

  

[Discussion]

• If you look at the gap between "win" and "lose draw", the graph on the far right is the best. You can see from the "losing draw" data of the number of balls captured that the number of balls captured within the enemy team is generally small. (Areas with fewer balls taken than winning matches are shown in blue.) There are many possible reasons.

Is it a combination of formations? Are you being turned behind because you have a counter target? Is the player who is pointing the cursor at the time of defense different from usual?

Next, let's take a look at "Ball Lost" in the figure below! Unlike "ball capture", there are some good results in "losing draw" games in some areas. What I was interested in was that the number of ball lost in the front half space (second row from the left) was increasing.

In Half Space, it is an important area where you can use both sides and the center, and you can also shoot by yourself. There are many possible reasons why the ball is lost despite the chance.

Build up with a through pass even though the enemy is close Press the dash button The choice of side or center is biased and read by the other party

We will make a hypothesis, interview the tester, get additional data such as videos, and verify it. It will lead to the next analysis.

Then next!

■ Multiple parameter analysis

[Graph]

• Graph multiple parameter information together! We selected four parameters: "number of passes", "pass success rate", "score", and "goal".

  

[Discussion]

• From the "Goals" in the figure on the right, you can see that only "pass success rate 80% or less" can be played in games with 0 points or less. On the contrary, the number of passes does not seem to affect it. Similarly, the "score" in the figure on the left shows that the higher the "pass success rate", the higher the score. The "goal" in the middle figure has an interesting result! Neither the pass success rate nor the number of passes seems to affect the number of goals scored.

For this tester, the result was that the pass success rate affects the attack power, but it does not seem to be related to the goal. The next analysis looks good to take:

How can I increase the pass success rate? What are the other parameters that affect the number of runs?

■ Tips for data analysis!

• The key to such data analysis is to understand the parameter dependencies / inclusion relationships! For example, the data "possession" is a parameter that comes out by accumulating a number of factors.

• Hmm ~~ a lot! !! Lol However, it is not enough to look at all the parameters in detail. ** "What particle size is the most efficient way to see the information you want?" ** It is important to search for a meaningful chunk of data.

■ Summary

• After all, rich graphs provide more valuable information than cheap diagrams such as the first histogram. I hope you read this article and think "I want to use it!" From the players. If you create something that doesn't match your needs, it's the same as garbage. We will continue to increase the value of the software with the cooperation of testers.

■ Future plans

• We are planning to expand the function. ・ Relationship between formation and each parameter ・ Creating a graph to grasp the amount of change in parameters (aiming to make weekly growth easier to see) ・ 100% character recognition accuracy using template matching

** As an extra edition ** ・ Automatic retweet of counter recruitment BOT

** Stay tuned for part3! !! ** **

■ To everyone who read

• ** If you are interested in my activities, please feel free to DM me on Twitter ♪ ** It's okay if it has nothing to do with the content of the article, such as "Can you please do this?"

■ Reference URL

• Draw a Scatter plot with matplotlib
• Implementation of various graphs by cheat sheet and Python for graph creation
• Perform face detection with OpenCV for all images in the [Python] folder
• seaborn jointplot
• Placing Colorbars
• Introducing 6 typical data analysis methods! Now you can do basic data analysis