Agustin Leperini's Projects
Text analysis app based on docker and bash with regex
Vehicle classification from images using transfer learning and fine-tuning of a ResNet50 convolutional neural network.
Determine whether a new loan applicant will be able to repay their debt or not. Manipulated and visualized data, performed data pre-processing for a very small dataset of 50,000 applicants. Trained many supervised models like Random Forest, Boosting ensemble learning with LightGBM, XGBoost and CatBoost, and Stacked ensemble learning with Soft Voting and Stacked models achieving +0.64 ROC AUC. Compared that result against a Deep Learning neural network like a Multilayer perceptron. Deployed in AWS instances using Docker and also using API-based web-service application with Flask.
Predict whether a person applying for a home credit will be able to repay their debt or not. Data pre-processing for a large dataset of +350,000 transactions. Trained many supervised models achieving +0.72 ROC AUC. Models used where DecisionTree, XGBoost and LightGBM.
Deploying an image classification ResNet50 model with an API-based web-service application with Flask, Docker and Redis.
Building a dataset by connecting to the NBA API and extract information from other sources to perform regression models to estimate players salaries and binary classification to predict All-NBA players selections.
Binary sentiment classification from Stanford AI Lab movie review dataset.
Projects, internships, university classes, courses to practice and become better every day. This is where lifelong learners live.