I was playing around with Amazon ML and built a quick hyperparameter optimization example based on Amazon's GitHub example for k-fold cross validation. I'm an engineer at SigOpt so there's a SigOpt example, but I've also included a non-SigOpt hyperparameter optimization pipeline that updates the old Amazon k-fold cross validation example to boto3, runs as a single file, and lets you provide a list of hyperparameters upfront.

This blog post is a followup to a talk at MLConf NYC. Hope it helps you optimize your hyperparameters!

t-SNE of TF-IDF of Hillary (blue) and Donald (red) tweets, created at a hackathon. Hover to see tweet text! Notice how tweets that are close in space share related ideas (blue cluster in top left quadrant is Spanish, for example). Learn more about t-SNE: http://scikit-learn.org/stable/modules/generated/sklearn.man... Election data GitHub repo: https://github.com/WiMLDS/election-data-hackathon

From a tech talk for SF Bayarea Machine Learning Meetup at Heroku, recorded by Hakka Labs. Second ever tech talk!

From a talk for the Women in Machine Learning and Data Science meetup on Hyperparameter Optimization

Hi! I'm the author, and I can answer questions if you have any.

Hi, I'm the author, and I've been working the RESTful API redesign for SigOpt. I've learned a lot about APIs over the last few months and I'm excited to share, and I'm always looking to learn more. Let me know if you have any questions/comments!

Yup, summary is about right. Being funny is a great point. Could probably expand on that in another post. I've heard before that you're not supposed to try to be funny at the last minute in a talk/essay if you're not funny in general, though, so that advice is a tricky balance. I think being funny and joking around during presentations comes from feeling comfortable with your audience.