Day 39: Python 3.7 Released

Stacking WaveNet autoencoders on top of each other leads to raw audio models that can capture long-range structure in music. Check out our new paper: https://t.co/JGHfu2OkAL

Listen to some minute-long piano music samples: https://t.co/CzF7PnxVC3 pic.twitter.com/snWRKHbnwr

— Sander Dieleman (@sedielem) June 28, 2018

Guided Evolutionary Strategies: Escaping the curse of dimensionality in random search, from Google Brain: @Niru_M @Luke_Metz @GeorgeJTucker @JaschaSD.
Paper: https://t.co/6UmZdNyzLz
GitHub: https://t.co/BrKeRlkEBS
Notebook: https://t.co/AzuBF7coAP pic.twitter.com/HIGEEi7FkN

— hardmaru (@hardmaru) June 28, 2018

This algorithm beats CMA-ES (kind of like the LSTM of the ES world) for a few tasks. CMA-ES (from Nicolaus Hansen) is still my algorithm of choice for blackbox optimisation. I wonder if this algo will consistently beat CMA-ES on a variety of different tasks and make me use it ..

— hardmaru (@hardmaru) June 28, 2018

The great thing about this paper is how they made the results and method completely open, and we can easily try it out using the code in the colab notebook on other tasks: https://t.co/5Vh9lyAwOK

— hardmaru (@hardmaru) June 28, 2018

Guided evolutionary strategies: escaping the curse of dimensionality in random search. A principled method to leverage training signals which are not the gradient, but which may be correlated with the gradient. Work with @niru_m @Luke_Metz @georgejtucker. https://t.co/LNPHDUrDFu

— Jascha (@jaschasd) June 28, 2018

perhaps it's too early to tell but i'm really impressed by the SNGAN w projection and glad i persevered through the Chainer install - restarted the training this morning due to a bug but the GAN converges quickly and results are diverse and nicely category specific. pic.twitter.com/XiR9Xe4V6c

— helena sarin (@glagolista) June 26, 2018

more details - on the same dataset (pretty modest) the other GANs either haven't converged or got to the mode collapse pretty quickly...
and also 1080ti made quite a difference - training on 128X128 res consumes more than 10G RAMhttps://t.co/SW45s8tACM

— helena sarin (@glagolista) June 26, 2018

This is the first story I worked on where I took advantage of Python's new Altair charting library during development.

It's already much better than the alternatives, and it's improving every day. Get on board. pic.twitter.com/0waCzZDEnc

— Ben Welsh (@palewire) June 27, 2018

Who kissed whom in #TheOffice (US). #datavizhttps://t.co/X3E2HFXmgB pic.twitter.com/329BqvEHVC

— Randy Olson (@randal_olson) June 27, 2018

Reproducibility tip of the day: If you're sharing research code & data, I'd recommend splitting it into three parts.

1. Raw data & preprocessing code
2. Preprocessed data & modelling code
3. Trained model & evaluation code

— Dr. Rachael Tatman (@rctatman) June 27, 2018

Debugging “Software 2.0” requires a bit more effort. How to unit test machine learning code: https://t.co/Hnq4TjYiq2 pic.twitter.com/pff03X9GcO

— hardmaru (@hardmaru) June 27, 2018

Article written by @TheNerdStation. They also wrote a simple framework for testing ML code: https://t.co/vQf82i6rxm

— hardmaru (@hardmaru) June 27, 2018

Are you interested in text as data for computational social science? You might be interested in my slides on using information-theoretic methods for text analysishttps://t.co/kMDSLMP6G1 pic.twitter.com/zGQTGqKTVT

— Ryan J. Gallagher (@ryanjgallag) June 26, 2018

Survival Analysis to Explore Customer Churn in Python https://t.co/A3Tb9RCHsd pic.twitter.com/IdUnLgqJu5

— KDnuggets (@kdnuggets) June 27, 2018

Liza Darrous did a great presentation on circular visualisations @RLadiesLausanne recently! I used jokergoo’s circlize package for a quick #tidytuesday spin. circlize can use tibble data directly but some formatting options are only for matrix. https://t.co/GBrouNj0je #rstats pic.twitter.com/iTTVAsLx1Q

— Xavier (@xvrdm) June 27, 2018

Next-level network analysis 📦:
"netrankr: Analyzing partial rankings in networks" ✍️ @d_mathlete https://t.co/BRZM3OcalI #rstats pic.twitter.com/D3m4l9YP71

— Mara Averick (@dataandme) June 28, 2018

Pyod - A Python Toolkit for Outlier Detection (Anomaly Detection). https://t.co/30LAaXVqt4 #python pic.twitter.com/dnTJLCmrEP

— Python Weekly (@PythonWeekly) June 27, 2018

#rstats source code formatter styler v1.0.2 is on CRAN. We don't try to style your non-R code chunks in R markdown anymore, refined styling around tilde, fixed some edge cases and more. Full changelog here: https://t.co/VTv6QRlmAe

— Lorenz Walthert (@lorenzwalthert) June 27, 2018

PyLogger - A singleton logger that will be used globally by the project https://t.co/dChbAdYiK2

— Python Trending (@pythontrending) June 27, 2018

Reinforcement Learning outperforms previous algorithms at RNA Design. Analysis of its solutions shows it has successfully learned some advanced strategies previously identified, allowing it to solve some very difficult structures. @PandeLab https://t.co/yZ4uO7jo9s

— Vijay Pande (@vijaypande) June 27, 2018

Python 3.7.0 is released! Bring out the celebratory libations. Thanks @baybryj and a cast of thousands on python-dev and GitHub. https://t.co/wEG4vO76Rd

— Guido van Rossum (@gvanrossum) June 28, 2018

Nice summary of the neat stuff in the just-released Python 3.7 at https://t.co/MhgkS6Yr2S . Lots of good stuff there, but I think the breakpoint() function is my favorite - makes the debugger "for the masses" rather than just those who know the magic incantations!

— Erik Tollerud (@eteq) June 27, 2018

Yes, but did it work? Evaluating variational inference https://t.co/AtNAhDoQV3

— Andrew Gelman (@StatModeling) June 27, 2018

📽 my slides from last night's Greater Boston useR meetup.
💬 "That's not [data] science!” https://t.co/t5zxsneVl8
Thanks again to @tcbanalytics, and @WeWorkBOS! pic.twitter.com/Vk8mAdP1Cq

— Mara Averick (@dataandme) June 27, 2018

PhD student @jovialjoy is one of @techreview's 35 Innovators Under 35 for her pioneering work fighting algorithmic bias in facial recognition and analysis technology. https://t.co/hjiGyHmWnx @AJLUnited

— MIT Media Lab (@medialab) June 27, 2018

The future is where you may lose 1 out of every 333 of your customers due to a suboptimal experience caused by a psychological experiment where neither the content creator nor viewer gave consent or received notice.
Isn't it nice to be a sentient cog in the big data machine? https://t.co/SdN2aMwy4F

— Smerity (@Smerity) June 28, 2018

The Gates Foundation didn't just fail on Education Reform, it actively harmed teachers. And it also undermined trust in the scientific method.https://t.co/aYkUoq2cm1

— Cathy O'Neil (@mathbabedotorg) June 27, 2018

YouTube content moderation: "Deploying machine learning actually means more people reviewing content, not fewer" https://t.co/pHVm4FnDcF -- yet again AI as complement to human labor.

— Nick Diakopoulos (@ndiakopoulos) June 27, 2018

Watched the Westworld Season 2 finale, and have to say it's great "food for thought" re: machine learning (esp in final S2 episodes... "testing for fidelity") and AI ethics. I recommend it if you like interesting shows with techie themes!

Heads up, though: it's very violent.

— Data Science Renee (@BecomingDataSci) June 27, 2018

Researching voices in the #AI + #ethics space. Who should I know about? Starting, emerging, unknown, known, experts, not-an-expert-but-interested, not-an-expert-but-really-you-are. Specifically seeking geographic + field of study diversity. Please RT. DM's open.

— amcasari (@amcasari) June 25, 2018