Closed dudung closed 5 months ago
Problems faced during building the project of this courses: The deep reinforcement learning method for optimizing stock portfolios is a relatively new field that not everyone is familiar with. Sometimes, I find it challenging to locate related literature. In deep reinforcement learning, the central — and most intricate — aspect of the process lies within its environment: it mimics the behavior of the system being modeled. The more detailed the environment, the more accurately we can replicate its behavior.
Technically speaking, I have found the most difficult part to be experimenting with new DRL libraries such as OpenGym, GymAI, and various deep learning frameworks like PyTorch and TensorFlow. These libraries have been updating rapidly, which often leads to outdated repositories when seeking others' work. This can be frustrating, as it detracts from focusing on refining technical skills once the core algorithmic concepts have been mastered.
There is no single "best" algorithm for the deep reinforcement learning process; it depends on the specific problem at hand. When discussing "the algorithm," we must also consider the architecture of the deep learning models and their parameters. The main idea of DRL lies in the integration of the algorithm with the DRL environment itself. Overall, delving into the development of deep learning theory and its applications in reinforcement learning has been a good experience for me.
P.S. I forgot to click the "presence" button in my SI-X account on March 6. Could you please fix it as I am still actively involved and doing my work in this course?
Thank you, Pak Dudung
Nama : Nita Dwi Fitriani Nim : 20922303
As a beginner in Machine Learning research to predict the Phonon Density of States (DOS), I faced some significant challenges. First of all, as someone with no background in materials science, understanding basic concepts such as DOS Phonons can be challenging. I had to learn the basics of solid physics and the nature of matter before I could understand the context required to perform DOS Phonon predictions.
The next challenge is to understand the appropriate Machine Learning algorithms and techniques for this problem. Although I had a basic understanding of programming, selecting and applying the right Machine Learning algorithms in the context of solid physics was new to me. Therefore, I searched for references to increase my understanding of this research.
Although the library of the method I used gave me full access to the data, they did not provide a way of reprocessing the data before it was processed with the data structure they used, so I had to generate new material data to test this method in order to validate and evaluate the model developed.
Name: Dewa Made Alit Adinugraha NIM: 20922006
Assignment 5 In this research on sentiment analysis there are several problems that I face, these problems occur not only this time, but I have faced several times before. There is a limitation of access in obtaining data, this is influenced by the restriction of the party who provides the data, the API provided by twitter provides a very limited access on the information to be taken, the amount of obtained data were also the problem.
The next problem is the selection of algorithms in data processing, the study field of NLP is still very broad in science, because one word cannot describe information that is always precise related to the intended purpose in a comment. the selection of the right algorithm will greatly affect the accuracy and prediction of the value of sentiment. but still limited to positive and negative sentiment, the value of accuracy is very unlikely to be of high value if looking for the value of sub-category such as whether the value of negative sentiment has the meaning of cyberbully or opinion on an object
From this problem, I solved it by changing the data source that initially used twitter to YouTube comments, where the YouTube API has fewer restrictions so that more information will be obtained, later this information will be more useful than what was obtained before, because in number it will be obtained more. Likewise the use of algorithms to process it, I will use SVM, because from the results of my studies, SVM works efficiently and has a high accuracy value