In this paper I show that the mlcausality method with kernel ridge regression outperforms rival nonlinear Granger causality algorithms. Greater accuracy is acheived at execution speeds that are 10+ times lower than competing methods.

I develop mlcausality, a Python library for identifying nonlinear Granger causal relationships using machine learning techniques. This library allows the user to use many different machine learning techniques, such as random forests, XGBoost, or SVR, to recover nonlinear Granger causal relationships.

Top 3% finish on a highly competitive Kaggle competition. Used an ensemble of 25 CatBoost models to identify whether an age-related condition occurs from data with anonymized features. Used a repeated stratified K-fold cross-validation strategy to tune CatBoost hyperparameters.

43rd place (top 3% finish) on a challenging Kaggle competition. The objective was to predict Parkinson's disease progression on 4 separate scoring criteria at 4 separate points in the future using protein and peptide data. My solution used a combination of numerous CatBoost and linear regression models with parameters tuned using GroupKFold cross-validation.

24th place (top 4%) on the efficiency leaderboard on a competitive Kaggle competition. The objective of the competition was to predict essay scores on 6 different metrics. My solution involved extracting embeddings from the DeBERTa model and predicting essay scores using SVR and CatBoost. This solution achieved excellent results without fine-tuning DeBERTa!

Co-created a hands-on deep learning course that uses Keras and Tensorflow. Features the development of Transformer models for sequential data and time series prediction; advanced convolutional neural networks (CNNs) using Keras; unsupervised learning, and more!

This is a hands-on course written for IBM's Skills Network platform intended to teach data science beginners the fundamentals of exploratory data analysis (EDA) and feature engineering. This hands-on project received over 850 enrollments in less than 4 months and obtained a review score greater than 4.5/5.

Learn how to fine-tune a transformer-based neural network using PyTorch. First, you'll train a neural network on the IMDB dataset (4 classes) from scratch, and then perform a full and last-layer fine-tuning on the AG News dataset (2 classes). Results will be compared to the performance of a model trained on AG News from scratch.

This PyTorch course provides learners with a solid understanding of the basics and core principles of PyTorch, specifically focusing on tensor manipulation, dataset management, and data augmentation techniques. This is a popular course with over 1200 enrollments and a rating of 4.6/5 at the time of writing. If you need a quick intro to PyTorch, this course is for you!