Automating Credit Earning in EVE Online with AI

Automating Credit Earning in EVE Online with AI

Table of Contents:

1. Introduction
2. The Inspiration: EVE Online and the Mini Game
3. The Initial Approach: Brute-Force and the Limitations
4. Introducing Machine Learning
5. Preparing the Data for Training
6. The SKLearn Library for Machine Learning
7. Understanding the MLP Classifier
8. Selecting the Right Parameters
9. testing and Training the Model
10. Saving and Loading the Model
11. Results and Conclusion

🔍 Introduction

Have you ever wondered how artificial intelligence can be used to automate tasks or even cheat at a game? In this article, we'll dive into the fascinating world of machine learning and explore how it was used to cheat at EVE Online, a popular online game. Join me as I take you through the journey of implementing a machine learning algorithm to play a mini game within EVE Online and earn credits.

🚀 The Inspiration: EVE Online and the Mini Game

EVE Online, a massively multiplayer online game, features a unique mini game called the Human Protein Atlas. This game aimed to crowdsource protein marker identification by integrating it into the gameplay. While traveling between star systems, players could contribute to the research project by playing this mini game. The challenge was to classify different protein markers Present in a given cell image. However, due to the complexity of the task, the researchers relied on humans rather than computers for accurate identification.

💡 The Initial Approach: Brute-Force and the Limitations

At first, the idea of brute-forcing the solution seemed reasonable. Extracting characteristics from the images, such as the percentage of green pixels, the percentage of blue with green, and more, would help identify the categories. However, with 27 categories to consider, the conditions quickly became complex and hard to manage. To overcome this challenge and truly understand machine learning, a different approach was needed.

✨ Introducing Machine Learning

Machine learning, particularly neural networks, offered a promising solution. Neural networks are regression matching algorithms that aim to find the parameters that yield the desired results. By training the network with actual data containing known answers, it becomes possible to automate the classification process. This article focuses on neural networks as they can handle multi-label classification problems like the one found in the Human Protein Atlas mini game.

🔧 Preparing the Data for Training

To train the neural network, a dataset with images and their respective categories was required. Initially, an AI was developed to play the categories randomly and Record the correct answers. Screenshots were taken to capture the correct answers, and a JSON data array was generated, containing the necessary information for training and testing the neural network. The data included processed characteristics like the percentage of green and red pixels.

📚 The SKLearn Library for Machine Learning

To implement the machine learning algorithm, the SKLearn library was utilized. SKLearn provides a comprehensive set of tools for classification, regression, clustering, and more. It offers user-friendly documentation, making it accessible for beginners. The MLP classifier, a type of neural network, was chosen for this project. By exploring the documentation and understanding the various parameters, including the solver, alpha, Hidden layers, and random state, the most suitable configuration for the classifier was determined.

🔬 Testing and Training the Model

Once the JSON data was prepared, it served as the input for training the MLP classifier. The model was trained using the data and tuned for optimal performance. The goal was to achieve the highest percentage of accurate classifications. The training process involved iterating through the data and computing various values to feed into the neural network. The model was then saved using the pickle library, allowing it to be loaded and used for classification without repeating the training process.

💯 Results and Conclusion

Although the results of the project did not surpass expectations, the developed AI managed to achieve over 50% accuracy in the Human Protein Atlas mini game. While the ambition was to achieve a higher percentage, the project showcased the power of machine learning in tackling complex tasks. Following the project's conclusion, the university even organized a competition on Kaggle to identify the best neural network for protein marker identification. This experience further enriched the learning process and highlighted the importance of continuous exploration in the field of machine learning.

✅ Highlights:

• Discover how machine learning can be used to automate tasks in games
• Explore the unique mini game in EVE Online, the Human Protein Atlas
• Understand the limitations of brute-force methods in tackling complex tasks
• Introduce the concept of neural networks and their applications
• Learn how to preprocess data for training a machine learning model
• Dive into the SKLearn library and its MLP classifier
• Experiment with different parameters to fine-tune the model
• Test and train the model using a JSON dataset
• Explore the results and draw conclusions from the project
• Embrace continuous learning and improvement in machine learning

FAQ:

Q: Can machine learning algorithms be used to cheat in games? A: Machine learning algorithms can be used to automate certain tasks in games, potentially gaining an advantage, but it depends on the game's mechanics and the complexity of the tasks involved.

Q: How accurate was the AI developed for the Human Protein Atlas mini game? A: The developed AI achieved a success rate of over 50%, which was enough to receive the highest reward in the EVE Online game.

Q: What are some recommended resources for learning more about machine learning? A: The SKLearn library documentation is an excellent resource for beginners, providing clear explanations and examples. Additionally, online courses and tutorials can provide in-depth knowledge and practical applications of machine learning.

Resources:

• EVE Online
• SKLearn Library Documentation
• Kaggle