Something about heteroskedasticity

Over four million homes were sold in 2024, with a median sale price of $407,500 (source). That’s $1.66 trillion of residential real estate transactions in 2024 alone! And yet, house prices are notoriously difficult to predict—just ask Zillow! In this post, I’ll talk about some of the challenges in pricing illiquid assets (like homes), highlight how those challenges impact model design, and ultimately develop a fairly sophisticated house pricing model. Along the way, we’ll touch on topics like heteroskedasticity, hierarchical modeling, and Bayesian regression.

It's [contraint: clever remark]!

Modern LLMs are impressive pieces of machinery, capable of feats across many domains that were previously thought to be unique signifiers of human intelligence. This power and generality comes from their incredible size, which allows LLMs to compress huge quantities of information and recognize highly sophisticated patterns.

Almost exactly as easy as it sounds

One of the most powerful arguments for incorporating deep learning models into your workflow is the possibility of transfer learning: using a pre-trained model’s latent representations as a starting point for your own modeling task. This can be particularly useful when you have a fairly small number of labeled examples, but the task in question is similar to a pre-existing model’s task. So how easy is it to do transfer learning with an LLM? As we’ll see, with HuggingFace’s transformers library, it’s actually quite easy.