Our project requires the machine learning model written in Python to be deployed on the server, so we need some compatible back-end framework to work with our model. We first thought about Django, but for our MVP, we only need a lightweight back-end framework. Thus, after consideration, we think Flask would be a better choice for our project. Flask has a large community, it has 2.3k stars on Github and has over 500 contributors.
Javascript has lots of frameworks that benefit both front-end and back-end web development. Since we plan to build our project features on the web, such as data visualization, we also included React.js, Node.js, and many other js frameworks in our stack. As for team collaboration, all of us are familiar with Javascript language, so there would be no impediment for code understanding.
Project Version Control
Git is the most common version-control system used by programmers worldwide. It allows our team to develop different features of the product simultaneously. We could track the changes in our product with git commands.
Besides providing code management functionality based on Git, it further provides project management features such as issues, code review, approves, release, and project panels. We use Github issues and code reviews frequently. We could also present our project with its Github-page feature.
Scikit-learning is a machine learning library that features various traditional algorithms, which could speed up our model development process greatly. Since our MVP development cycle is relatively short, we would like to apply the standard algorithms such as PCA, Gaussian Mixture, Random Forest first to the outcomes. This library also inter-operates with another chosen stack Numpy.
Numpy is essential for data science no matter what library to use, as long as it is based on Python. The library is for large-scale scientific computing, it takes significantly less amount of memory when involves list operations, such as n-dimensional array calculation, array indexing, etc. Since our project will involve a lot of large scale computing, Numpy would be indispensable.
We use Pandas to organize the UBA dataset into easy-to-use data structures. Features are grouped into columns with Pandas, and then various arithmetic operations could be done between columns. The data structure is easy to read, any rows or columns could be extract t to see the value. This is especially handy when we do initial exploratory data analysis to find potential correlations.
Communication We have chosen two tools for our team communication.
We choose Slack since all of us are familiar with this communication tool. We have a private channel for our team Sphinx for text messages. We added Github apps inside our private channel for repo update notifications. Furthermore, we could contact the subject matter experts within the workspace DCSIL directly for the issues we meet.
We use Microsoft Teams for virtual meetings for its fast connection speed. In addition, the call feature in Slack is a paid feature, and we could have virtual meetings and share screens for free in Microsoft Teams.
We choose Python for our machine learning model development for several reasons.
Reason 1
The first reason and is also the major one, Python is no doubt the most popular language for machine learning and data science. There are numerous powerful libraries and frameworks built on top of Python.
Reason 2
Python code is easy to understand. When involving all sorts of statistical equations, readability is all-important.
Reason 3
Our project involves training with a huge amount of data, GPU acceleration would be necessary. Speed up model written in Python on the CUDA platform is relatively convenient.
Generalist Engineer • 1 stack
jerryishihara demonstrates balanced experience across multiple technical domains with versatile stack choices.