NCU machine learning course final project
-
Motivation
At first, we wanted to build a model to predict stock based on the analysis of ptt users' discussion. After some research, we found that idea is somewhat difficult for us. But fortunately, we discovered an interesting data set on Kaggle which is about the start-ups. Also, it's related to our original idea about predicting prospect of a stock/company, so we came up with a new plan based on this data set.
-
Goal
- Provide a interactive and fun way to know about start-ups.
- Give some possible invest targets.
- Find out the key traits of a company to be successful.
- Passing the class.
“Unicorn”, Aileen Lee called a successful company, refers to a privately held startup company with a value of over $1 billion.
Our work is to know the potential of this company whether it could become a “Unicorn” company or not, in the other word, the probability of becoming a “Unicorn” company will be our target value.
The feature of explanatory value used for fitting the model will be company’s funding date, industries, financial, people, technology.
In order to gain these features ,we consider to use the data from CrunchBase,a web database about start-ups, including various type of data such as founder, finance, and investors, etc.
The original source of the data set we use is from here.
- Investors that are seeking a practical way to know if a start-up will become a unicorn.
- People that are seeking the chance to enter a start-up.
- Startup founders who want to evaluate the their own or other's company by objective index(s).
從crunchbase取得各式新創公司的包含名稱、地理位置、資金、營收等資訊。
Algorithm to Predict the Future Development of Start-up Companies
UI to Interact with NT-D
- Features
- Show the possibility if a start-up will become a unicorn.
- Python 3.10.8 powerful programming language, able to use the package we need.*
- PyTorch PyTorch has tensor computation (like NumPy) with strong GPU acceleration to shorten our training time. DNN we used will be built on a tape-based autograd system. NumPy, SciPy, and Cython are all able to extend PyTorch. PyTorch also has multiprocessing which is useful for data loading and Hogwild training.
- PyTorch Lightning PyTorch Lightning can make the PyTorch model easier to build, connect, optimize.
