My blog

This is mostly a home for simulations for my questions and answers on stats.stackexchange.com and stackoverflow.com.

Here also lies a list of my contributions to open source software.

dir/file link	q/a link
select_on_test.ipynb	Demonstrate that a model can simultaneously be selected and evaluated on a test set
train_on_test_features	For high rank data and a small test set, train a PCA on test set features to boost test set performance!
precision_drop.ipynb	A simple answer to: why did precision drop in production?
auprc.ipynb	Demonstrate that integral approximators are trying to hurt you
db_sampling_rate.ipynb	Calculate a sampling rate for a database query
negative_vs_downsampling.ipynb	(not done) What's the need to formulate negative sampling for contrastive training?
to_batch_or_not_to_batch	(bad!) Mathematically analyze and demo dynamic batching
var_pred_var_error	Does higher variance in predictions result in higher variance error estimation?
sample_via_gumbel	Demonstrate that one can sample directly in log-space
(external) cappr	Demonstrate that a more usable version of zero-shot text classification works
langchain_save_all.ipynb	Save all method calls. Inspired by this issue

My dumber code dumps are in dumpy.

Setup

Need Python 3.8+

Create an environment blog using venv:

cd /your/venvs

python -m venv blog

source blog/bin/activate

python -m pip install -r /path/to/blog/requirements.txt

If the notebook says that it needs to run on a GPU machine, and you have a Google account, open the notebook in Google Colab.

Usage

Interact w/ the code via Jupyter. I like VS code notebooks.