The topic of how to create and maintain reproducible experiments has come up a few times recently. I've previously discussed ways to do reproducible hyper parameter sweeps and as part of that briefly discussed using anaconda environments. This (short) post elaborates on that with what I've learned since then. I'll …
Conducting a thorough literature review for a new idea is a critical yet difficult research skill to learn. Even now, for areas I'd like to think I have a good grasp on, its difficult not to miss something important; multiple times, I've done a literature review only to find a …
AI2 recently released version 1.0 of its excellent AllenNLP library.
The goal of this (living) blog post is to list code and configuration changes I've made in my own code to migrate from 0.9.0 to 1.0.
While I was writing this, I also found that the …
In my last post on parameter sweeps, I mentioned that I found a nice way to combine AllenNLP and comet.ml.
In this post, I'll share my code that does this via the (experimental) callback trainer.
The 1.0 release of AllenNLP is currently actively in development and the callback …
In the past couple weeks I've been working on writing machine learning code in python with the following goals:
Developing new machine learning code is often error prone and takes many iterations of the write-run-debug loop. In this context, I specifically refer to saving time fixing errors that crash the program--not than those that cause models to be incorrect in subtle ways (for that, see Andrej Karpathy's blog post …
We have updated our project site for qanta.org with many more resources.
The results of our NIPS 2017 Competition are now part of The Springer Series on Challenges in Machine Learning book. Our chapter, Human-Computer Question Answering: The Case for Quiz Bowl, is now available.
Summer 2017 I was a data science intern at Riot Games. It was an amazing experience for many reasons, and I'm excited to share the video that Riot Games produced featuring interviews with all the data science interns!
More about the internship, and applying at the …
Description: Intuitive introduction to what the Kl divergence is, how it behaves, and why its useful
The KL divergence is short for the Kullback-Leibler Divergence discovered by Solomon Kullback and Richard Leibler in 1951. Semantically, divergence means the amount by which something is diverging, and diverging in turn means to lie in different directions from a different point. In the case of KL divergence we …
