v-trungdt21/svm-from-scratch

Build SVM from scratch using only Python and some helper libraries: pandas, numpy,..

Support Vector Machine from scratch

Build SVM from scratch using only Python and some helper libraries: pandas, numpy,..

There are 5 tasks in total:

• Task 1: Inference.
• Task 2: Kernelize SVM.
• Task 3: Training using convex optimization library.
• Task 4: Training using self-implementeed library.
• Task 5: Visualization + Comparison with sklearn.

Installation guide 🔥

$ git clone https://github.com/trungdt21/svm-from-scratch
$ cd svm-from-scratch

# if you use pip
$ pip install -r requirements.txt
# if you use conda
$ conda create -n svm --file package-list.txt

$ pip install -e .
$ pre-commit install

Currently used hooks:

• isort: Sort imports.
• black: General python formatter.
• flake8: Check PEP8 convention.
• pre-commit: Run those hooks before commits.

Work Convention ✨

• Code convention: Follow PEP8 convention (include docstrings for functions).
• Update __init__.py when you add new modules.
• Use pre-commit so that the code would not be messy when merge between commits:
  • Usually, you can just git commit and pre-commit will run itself.
  • If you want to run it manually, use pre-commit run --all-files to run all files, pre-commit run <hook_id> to run a specific hook.
• Remember to put understandable commit descriptions.
• Update taskX.md of your task (documentation, progress, API documentation) usually so that other can follow it.
• Write tests if you can.

API

Install

  # Install the package
  pip install -e .

Usage

from svm import SVM_SMO, SVM_cvxopt

# We support 4 kernels: linear, sigmoid, poly, rbf.
model = SVM_SMO(C=100, kernel="linear", degree=3.0, gamma=1.0, coef=0.0)
or
model = SVM_cvxopt(C=100, kernel="linear", degree=3.0, gamma=1.0, coef=0.0)

model.fit(X,Y)
model.predict(Xtest)

Result

Contributors 🥺

• Phan Hoang Viet
• Tran Duc Nam
• Nguyen Hien Tuan Duy
• Dao Tuan Trung
• Mentor: Dzung Nguyen