Posts

Derivation of Glorot initialisation

Filling in the details of Glorot’s original derivation of his eponymous initialisation method, introduced in his landmark 2010 paper and which is now the default method for initialising the weights of deep neural networks.

2024-03-08

Machine learning Deep learning Neural networks

The German tank problem

Estimating the upper bound of a discrete uniform probability mass function from sampling without replacement using frequentist and Bayesian techniques.

2024-02-24

Probability Statistics

Name conventions for temporal data columns

For goodness' sake, can we please all agree to some standard naming conventions for columns containing temporal data?

2023-12-08

Data engineering Data cleansing Best practices

How I evaluate machine learning models

Description of the solution I use in day-to-day data science work to evaluate my machine learning models.

2023-11-03

Machine learning Model evaluation

A dense percentile rank function in Snowflake

Implementation of a 'dense' percentile rank function in SnowSQL.

2023-10-27

Snowflake SQL

A foreshadow of my data science career in my engineering undergrad: Kriging

AKA Gaussian process regression for the geosciences.

2023-10-06

Gaussian process regression Statistics