peter frick

ML performance is a subset of ML business value. Encoding business goals into the objective function of models boosts overall impact @MLconf #mlconf

Interesting explanation of variational autoencoders as a generalization of LDA, and how that’s used to scale recommendations @MLconf #mlconf

The area under the ROC curve (AUC) is so frequently criticized and misunderstood that I often wonder whether I am the metric’s only fan. Let me explain why how I see and value the AUC. (thread)

A nice fact I like: Every matrix corresponds to a graph, and so familiar things (e.g. matrix multiplication) have nice pictures! Another nice fact: joint probability distributions *also* correspond to graphs. They have telling pictures, too. New blog post! math3ma.com/blog/matrices-…

@Damienrj @_alexrhee @peclevenger Anyone else reading this?

Q: what do you do for a living? A: create bugs while fixing other bugs.

A comprehensive overview and break-down of 14 NLP research paper highlights from 2018

Goodbye, mid-range shot. flowingdata.com/2019/01/15/goo…

A recent paper in NEJM shows that children born in August are 30% more likely to be diagnosed with ADHD than children born in September, probably because they start school younger. I reanalyzed the data using Bayesian logistic regression: allendowney.com/blog/2018/12/0…

@rctatman Imagine predicting what I’m like based on a sibling. He is a model of me. If you don’t learn about him to predict me, you are excluding useful info (bias) But if you learn everything about him and apply it to me thats using too much info since we are different ppl (variance)

Got a recent question about this & didn't have a great answer: do y'all have any favorite resources for helping build a good intuition around bias/variance trade-off?

Congrats Parker!

Slides for my talk @py_bay about AI/ML and Python are located here: speakerdeck.com/teoliphant/ml-… #PyBay2018 #MachineLearning #Python @NumFOCUS @TensorFlow @PyTorch @ChainerOfficial @ApacheMXNet

Clicked for the title, stayed for the memes

RT @afshinea 🔥Set of illustrated Machine Learning cheat sheets from Stanford's CS 229 class: Deep Learning: stanford.io/2BsQ91Q Supervised Learning: stanford.io/2nRlxxp Unsupervised Learning: stanford.io/2MmP6FN Tips and tricks: stanford.io/2MEHwFM

How to Create an ARIMA Model for Time Series Forecasting in Python machinelearningmastery.com/arima-for-time…

One of the top questions we've been getting from new Python coders is "how should I structure my Python projects?" With this tutorial, we want to give you a dependable Python application layout reference guide that you can refer to: realpython.com/python-applica…

how to train a neural network for text classification using keras #neural-networks-text_classification" target="_blank" rel="nofollow noopener">frickp.github.io/neural-network…

"Advances in Semantic Textual Similarity": @GoogleAI's new post on universal text embeddings: ai.googleblog.com/2018/05/advanc… Pre-trained 512 dim #Tensorflow model: tensorflow.org/hub/modules/go… (already available in @scaletext_ai)

"Context is Everything: Finding Meaning Statistically in Semantic Spaces." A replacement of tf-idf that is actually better than tf-idf (finally?!) arxiv.org/abs/1803.08493

virtual memory

Just for fun... Here's how an outfield of all Buxton would compare to an outfield of all Pujols based on 2017 sprint speed. Each colored band is a second of hang time. outer band is 6 seconds, inner band 3.