At start of 2016, when had started my ML related education I created portfolio repository, named MyRoadToAI, where I kept uploading new exercises or mini projects. As some time have passed I'd like to restart this idea and increase the quality of my work.
The project is a showcase how to create ETL Pipeline and Machine Learning Pipeline in order build NLP model capable of attaching any of 35 categories to the message. Interaction with model can be done via website hosted on Flask server.
The goal of this project is to present the capability of gathering and analyzing data about a specific topic (in this case, Dota 2 player scene). It consists of ETL process where data is downloaded, cleaned, and transformed, as well as the data analysis process. The output of the project is a blog post communicating lessons learned.
System assembled from three models: human detector, dog detector and dog breed detector. For dog images it return most similar dog breed name out of 133 possible. For human it says to which dog breed human image is most similar to.
Example project showing how communication between Flask and Gunicorn can look like. Projects hosts ensemble of XGBoost and Neural Network models for MNIST images classification.