youtube_recommender_ML_app

A E2E project for creating a video recommender app based on YouTube videos.

Project NOTES:

• As final project from course provided by Kagglemaster Mario Filho, the target of this project is to implement a bunch of shortcuts and best practices learned in his course.
• From data extraction to deploying model, the project uses an ensemble model (LightGBM and RandomForest) to predict which ytube video would be interesting based on three keywords ["machine+learning", "data+science", "kaggle"].

Project link: https://mltubeapp.herokuapp.com/

Check below a high-level diagram of the project:

Project next steps:

• Data Extraction
• EDA
• Modeling
• Simple Front Page
• Heroku Deploy
• API Prediction
• Data Extractor broken! - Fix Data Extractor
• Implement some Database
• Implement Cloud of words chart based on TFIDVectorizer

Packages used:

• gunicorn==20.0.4
• Flask==1.1.2
• requests==2.25.0
• beautifulsoup4==4.9.3
• pandas==1.1.4
• joblib==0.17.0
• numpy==1.19.4
• scipy==1.5.4
• scikit-learn==0.23.2
• lightgbm==2.3.0

Running project:

Running local with Flask:

1. Git clone repository
2. Create venv
3. Activate venv
4. Install requirements
5. Start Flask server

git clone https://github.com/fduque/youtube_recommender_ML_app.git ytube
cd ytube
python -m venv .ytube 
source .ytube/bin/activate
pip install -r requirements.txt
flask run

Running local with Docker:

1. Turn on Docker Desktop
2. Execute build docker
3. Run docker

docker build . -t appinstance1
docker run -e PORT=80 -p 80:80 appinstance1

Related links:

• https://github.com/ledmaster