Guilherme's Blog

My thoughts and experiments on machine learning and related stuff

clustering   causal inference
Diagnosing validity of causal effects on decision trees

Diagnosing validity of causal effects on decision trees

Diagnosing confounding on leaf nodes of decision trees

classification   dreduction
Solving the mystery of my dog's breed with ML

Solving the mystery of my dog's breed with ML

Using the Stanford Dogs Dataset, deep learning, and explainability through prototypes to infer the unknown breed of my dog

classification   engineering
Training models when data doesn't fit in memory

Training models when data doesn't fit in memory

Using Dask and some other tricks so you can train your models under memory constraints

clustering   dreduction   causal inference
The Bayesian Bootstrap

The Bayesian Bootstrap

Faster, smoother version of the bootstrap that yields better results on small data

clustering   dreduction   causal inference
Calculating counterfactuals with random forests

Calculating counterfactuals with random forests

Another tree-based causal inference model using nearest neighbors on the embedding produced by a Random Forest

clustering   dreduction   causal inference
Calculating counterfactuals with decision trees

Calculating counterfactuals with decision trees

Decision Trees can be decent causal inference models, with a few tweaks

bayesian   regression
Risk and Uncertainty in Deep Learning

Risk and Uncertainty in Deep Learning

Building a neural network that can estimate aleatoric and epistemic uncertainty at the same time

bayesian   bandits   regression
A Practical Introduction to Randomized Prior Functions

A Practical Introduction to Randomized Prior Functions

Understanding a state-of-the-art bayesian deep learning method with Keras code

clustering   dreduction   causal inference
A Causal Look At What Makes a Kaggler Valuable

A Causal Look At What Makes a Kaggler Valuable

Using causal inference to determine what titles and skills will make you earn more

clustering   dreduction   causal inference
Causal inference and treatment effect estimation

Causal inference and treatment effect estimation

Estimating counterfactual outcomes using Generalized Random Forests, ExtraTrees and embeddings

classification
Calibration of probabilities for tree-based models

Calibration of probabilities for tree-based models

Calibrating probabilities of Random Forests and Gradient Boosting Machines with no loss of performance with a stacked logistic regression

clustering   dreduction
Supervised dimensionality reduction and clustering at scale with RFs with UMAP

Supervised dimensionality reduction and clustering at scale with RFs with UMAP

Uncovering relevant structure and visualizing it at scale by partnering Extremely Randomized Trees and UMAP

bandits   bayesian
Risk-aware bandits

Risk-aware bandits

Experimenting with risk-aware reward signals, Thompson Sampling, Bayesian UCB, and the MaRaB algorithm

bandits   bayesian
Approximate bayesian inference for bandits

Approximate bayesian inference for bandits

Experimenting with Conjugate Priors, MCMC Sampling, Variational Inference and Bootstrapping to solve a Gaussian Bandit problem

bandits   bayesian
Non-stationary bandits

Non-stationary bandits

Solving a Bernoulli Multi-Armed Bandit problem where reward probabilities change over time

clustering   dreduction
Supervised clustering and forest embeddings

Supervised clustering and forest embeddings

Using forests of randomized trees to uncover structure that really matters in messy data

bandits   bayesian
Bootstrapped Neural Networks, RFs, and the Mushroom bandit

Bootstrapped Neural Networks, RFs, and the Mushroom bandit

Bootstrapped Neural Networks and Random Forests for solving a more realistic contextual bandit problem

bayesian
Thompson Sampling, GPs, and Bayesian Optimization

Thompson Sampling, GPs, and Bayesian Optimization

Mixing Thompson Sampling and Gaussian Processes to optimize non-convex and non-differentiable objective functions

bandits   bayesian
Thompson Sampling for Contextual bandits

Thompson Sampling for Contextual bandits

Solving a Contextual bandit problem with Bayesian Logistic Regression and Thompson Sampling

bandits   bayesian
Introduction to Thompson Sampling: the Bernoulli bandit

Introduction to Thompson Sampling: the Bernoulli bandit

Introducing Thompson Sampling and comparing it to the Upper Confidence Bound and epsilon-greedy strategies in a simple problem

kaggle
My first (and only) Kaggle Top 10%

My first (and only) Kaggle Top 10%

What I've learned and how I succeeded in a big Kaggle competition

A blog for sharing my thoughts and experiments

Very happy to start a blog where I share my ideas with the community