Peer Review for proposal - Group C

[GitHub-Traveler]

Overall, I think this project is very interesting as it incorporates machine learning algorithms. This could be very useful in the context of teaching algorithms in machine learning in a much more intuitive way, which I really appreciate. Some of my feedback/ideas are:

• I have not seen a dataset included in this project report. Due to the generality of the project, such dataset may not be needed, but I think some toy/examples datasets could be included to serve as an example in this project, such as make_moons() and make_circles(). The inclusion of tools to make random datasets to visualize is a great feature in my opinion.

• I think there should be a bit more features in this project, as this can be done really easily in scikit-learn library of python, and there is also TensorFlow playground for visualizing the neural network learning process and results.

• I suppose that your group wants to visualize the dataset and the process of learning the decision boundary (?) or something similar in this project. In that case, how do you intended to visualize the process of learning with datasets having larger than 3 features, and the decision boundary is a hyperplane? If it is just the process of making machine learning easy for layman, I think it lacks a bit in term of visualizing contents.

[thaohp113]

Thank you so much for your recommendations.

• Dataset: We’ll think of finding some toy datasets for self exploration.
• Features: We will have more features like the optimization algorithms if the time allow. Currently, we’ll prioritize building other components first. Our main core value lies in making ML inclusive and demonstrating a simple pipeline with simple algorithms first.
• Regarding high dimensional data, we think we can perform the algorithms implicitly and then visualize the compressed version with T-SNE method on a small dataset because we don't have any running GPU.

[tienvu95]

Thanks group C for the comment and Thao for the response.

One of the most important thing we need to consider is to explain how our product offers something new. Depends on the range of the ML algorithms we expect to visualize, we can pick a few toy datasets. The dataset that you use might not be as important as the clarity of your visualizations used to explain each algorithm.

The last point might be sth you can think about. If we don't want to make it extremely complicated, how can we make visualizations for 3+ dimensions easy to comprehend and human readable (introduce the concept of PCA? or pairwise plots for small dataset might be enough?)

Note that you don't have to reply to this comment, this is just my thought on what you can do to improve your project.