| 摘要: | 近年來人工智慧的發展迅速，Pytorch 為重要的深度強化學習框架之一，使用者可以運用軟體庫裡的強化學習運算法，簡單開始架設模型，並且訓練這個架設的模型，最後運用它來預測一些目標，簡單來說概念是從過去的資料中探討未來的發展性。我們認為，這項技術應用在學校系統上也一定能夠有實質的幫助。本文主要介紹如何將 DQN技術應用於分析學生成績分析資料上，再來下一步建立一個由第一學期時，必修課成績及必修課程為基礎，來選擇接下來學期的選修課，目標是從各個系所的眾多選修課程中，挑出容易使修課學生獲得高分的課程，此 DQN 模型目的是幫助學生在選課上有更多的資訊。
本文先使用 R 語言做資料前處理，針對所有大學部學生的課號及修課成績資料，首先將這些大學生每一學期的成績資料逐一排列，並選出人數最多的系所進行模型的參考訓練目標。再來是利用Python 加上 DQN 演算法來訓練模型，並調整模型內的參數，逐步訓練後找出最「DQN選課」的模型。雖然在本文最後有成功建立出目標模型，但由於此模型只是經由過去的數據來預測未來，並沒有實質上的模擬是否可行，所以還需要投入更大量的時間去證明它是否可用於現實的選課。
最後引用Deep Reinforcement Learning and Simulation as a Path Toward Precision Medicine 這篇論文，使用模擬的方式來去驗證Deep Reinforcement Learning 的實際可行性，並且將它投入精準醫療的領域，實驗最後顯示此方法的優點為，不僅可以探索臨床實踐和可用數據之外新的治療策略，還能隨著時間的推移，患者的個體疾病進展將捕捉到患者之間的差異，和單個患者內疾病進展的內在隨機性。
;In recent years, artificial intelligence has developed rapidly. Pytorch is one of the important deep reinforcement learning frameworks. Users can use the reinforcement learning algorithm in the software library to simply start to build the model, train the built model, and finally use it to predict some goals. In simple terms, the concept is to explore the future development from the past data. We believe that the application of this technology to the school system will definitely help. This article mainly introduces how to apply DQN technology to analyze student performance analysis data, and then establish a compulsory course based on the first semester scores and compulsory courses to choose the elective courses for the next semester. The goal is to select the elective courses for the next semester. Among the many elective courses, select the courses that are easy for students to get high marks. The purpose of this DQN model is to help students have more information in the elective courses.
This article first uses the R language to do data pre-processing. For the class numbers and class scores of all college students, first arrange the scores of these college students each semester one by one, and select the reference training target of the model conducted by the department with the largest number of students. Next, use Python plus the DQN algorithm to train the model, adjust the parameters in the model, and gradually train to find the most "DQN course selection" model. Although the target model was successfully established at the end of this article, because this model is only predicting the future based on past data, and there is no actual simulation whether it is feasible, it needs to invest a lot of time to prove whether it can be used for realistic course selection.
Finally, I quoted the paper Deep Reinforcement Learning and Simulation as a Path Toward Precision Medicine, using simulation to verify the practical feasibility of Deep Reinforcement Learning, and put it into the field of precision medicine. The experiment finally showed that the advantages of this method are not only New treatment strategies beyond clinical practice and available data can be explored, and over time, the patient’s individual disease progression will capture the differences between patients and the inherent randomness of disease progression within a single patient. |
