{ "id": 361838, "title": "Rocket league has a frenetic pace and why it's so hard to predict", "url": "/competitions/tabular-playground-series-oct-2022/discussion/361838", "author_name": "Mateus Coelho", "post_date": "2022-10-24T03:32:49.015000", "votes": 6, "comment_count": 0, "views": 0, "content": "

In order to better understand the game and create new features, I've spent some in Twitch watching Rocket League. What became very evident for me is the frenetic pace of the game. Just watch this video for 1 minute and you will understand. The ball goes all over the field so fast and the cars have to be constantly adjusting their positions.

\n

But how can we measure this frenetic pace in terms of our data? First let's look at the velocity.

\n

\"\"

\n

The average ball velocity is 40 units and the median player velocity is 31 units. That's quite high if we remember that the field is approximately 160x220 units.

\n

But how does that relate to our hability to predict the goal? I run a LGBM model in a 5-Fold CV setup so that I have a prediction for each row in the training set. The graph below shows the average prediction of each second before the goal (end of the event).

\n

\"\"

\n

Notice that until 1 second before the goal the model still not very sure if there's going to be a goal or not. Only in the last moment the average prediction peaks in 0.6.

\n

So, what can we conclude? Rocket League is frenetic game and somewhat unpredictable.

", "messages": [ { "id": 2001433, "postDate": "2022-10-24T03:32:49.017Z", "content": "

In order to better understand the game and create new features, I've spent some in Twitch watching Rocket League. What became very evident for me is the frenetic pace of the game. Just watch this video for 1 minute and you will understand. The ball goes all over the field so fast and the cars have to be constantly adjusting their positions.

\n

But how can we measure this frenetic pace in terms of our data? First let's look at the velocity.

\n

\"\"

\n

The average ball velocity is 40 units and the median player velocity is 31 units. That's quite high if we remember that the field is approximately 160x220 units.

\n

But how does that relate to our hability to predict the goal? I run a LGBM model in a 5-Fold CV setup so that I have a prediction for each row in the training set. The graph below shows the average prediction of each second before the goal (end of the event).

\n

\"\"

\n

Notice that until 1 second before the goal the model still not very sure if there's going to be a goal or not. Only in the last moment the average prediction peaks in 0.6.

\n

So, what can we conclude? Rocket League is frenetic game and somewhat unpredictable.

", "rawMarkdown": "In order to better understand the game and create new features, I've spent some in Twitch watching Rocket League. What became very evident for me is the frenetic pace of the game. Just watch [this video](https://youtu.be/Su_SV29obhE?t=7653) for 1 minute and you will understand. The ball goes all over the field so fast and the cars have to be constantly adjusting their positions.\n\nBut how can we measure this frenetic pace in terms of our data? First let's look at the velocity.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1844874%2Fc801b153023a7e7e687bb2e66e341186%2Fvelocity.png?generation=1666582295913669&alt=media)\n\nThe average ball velocity is 40 units and the median player velocity is 31 units. That's quite high if we remember that the field is approximately 160x220 units.\n\nBut how does that relate to our hability to predict the goal? I run a LGBM model in a 5-Fold CV setup so that I have a prediction for each row in the training set. The graph below shows the average prediction of each second before the goal (end of the event).\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1844874%2F13ff9e365dbc61ceec415fc7938b7090%2Fpreds_by_event_time.png?generation=1666582310016360&alt=media)\n\nNotice that until 1 second before the goal the model still not very sure if there's going to be a goal or not. Only in the last moment the average prediction peaks in 0.6.\n\nSo, what can we conclude? Rocket League is frenetic game and somewhat unpredictable.", "votes": 6 } ], "comments": [], "raw_markdown_by_id": { "2001433": "In order to better understand the game and create new features, I've spent some in Twitch watching Rocket League. What became very evident for me is the frenetic pace of the game. Just watch [this video](https://youtu.be/Su_SV29obhE?t=7653) for 1 minute and you will understand. The ball goes all over the field so fast and the cars have to be constantly adjusting their positions.\n\nBut how can we measure this frenetic pace in terms of our data? First let's look at the velocity.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1844874%2Fc801b153023a7e7e687bb2e66e341186%2Fvelocity.png?generation=1666582295913669&alt=media)\n\nThe average ball velocity is 40 units and the median player velocity is 31 units. That's quite high if we remember that the field is approximately 160x220 units.\n\nBut how does that relate to our hability to predict the goal? I run a LGBM model in a 5-Fold CV setup so that I have a prediction for each row in the training set. The graph below shows the average prediction of each second before the goal (end of the event).\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1844874%2F13ff9e365dbc61ceec415fc7938b7090%2Fpreds_by_event_time.png?generation=1666582310016360&alt=media)\n\nNotice that until 1 second before the goal the model still not very sure if there's going to be a goal or not. Only in the last moment the average prediction peaks in 0.6.\n\nSo, what can we conclude? Rocket League is frenetic game and somewhat unpredictable." } }