"The data were not normally distributed and therefore we used the Mann-Whitney U test, to test for....."
— Maarten van Smeden (@MaartenvSmeden) February 26, 2019
*don't say it*
*don't say it*
*don't say it*
*don't say it*
*don't say it*
*don't say it*
*don't say it*
"... a difference in medians"
*silent scream*
This is why systems like GPT-2 will be very effective at influencing discourse in places where close reading is not the norm (e.g. many social media discussions of politically charged topics) https://t.co/wRepSV34mX
— Jeremy Howard (@jeremyphoward) February 26, 2019
5/ Nowadays, all my personal projects start out as Jupyter notebooks. At work things are a bit trickier, but my ideal: 95% of my development time is spent in notebooks, and 5% spent copying shared code from notebooks to infrastructure.
— Eric Jang (@ericjang11) February 25, 2019
3/ and yet, he completely misses the point of why notebooks have unlocked so much productivity in data science, ML. Notebooks are the only IDEs that support interactive visualization out of the box. I always code visualization pipelines in notebooks before transferring to libs
— Eric Jang (@ericjang11) February 25, 2019
A great find within the PetFinder discussion threads — '9 General Tips for Kaggle' by ryches |Tip #1: Be OK with not knowing everything... https://t.co/utGMYXxicp pic.twitter.com/ejDYqgCEeM
— Kaggle (@kaggle) February 25, 2019
I thought this thread had good things, but was quite confusing. My summary: (smart) people don't like to be micro managed, but need clarity in goals, strategy, and ways to measure success. And, yes, context matters for all the above as I wrote here: https://t.co/hOd1wZqXO6 https://t.co/ia7K6oGnnf
— Xavier🎗🤖🏃 (@xamat) February 25, 2019
>>> a = np.zeros(5, 5)
— Andrej Karpathy (@karpathy) February 24, 2019
>>> TypeError: data type not understood
# thanks numpy, that's very helpful. pretty sure if isinstance(dtype, int): print("did you mean to use a tuple for size?") would have saved the human race a gajillion hours.
This is smart and on point from @l2k https://t.co/nfojHHDLgN
— dj patil (@dpatil) February 24, 2019
The #US now spends more than $11,000 per person for healthcare, while the #UK ~$4,000. Yet life expectancy has declined in the #US for each of the past 3 years, while it continues to increase in the #UK (graph adopted from @OurWorldInData) pic.twitter.com/rBPCsuGMpz
— Eric Topol (@EricTopol) February 24, 2019
"We define Internal Covariate Shift as the change in the
— Roger Grosse (@RogerGrosse) February 24, 2019
distribution of network activations due to the change in
network parameters during training." -batch norm paper
Those who cite ICS as a nebulous concept, can you give a plausible interpretation besides the obvious one?
"Fakery is gushing in from everywhere and we’re drowning in it. Deepfake videos. Easy-to-use software can generate audio or video. Even easier? Fake clicks, fake social media followers, fake statistics, fake reviews." https://t.co/r8yQLgzWij
— Rachel Thomas (@math_rachel) February 22, 2019
Reproducible Benchmarking of Keras and PyTorch Models https://t.co/Lmksa6nKPF #deeplearning #machinelearning #ml #ai #neuralnetworks #datascience #keras #pytorch
— PyTorch Best Practices (@PyTorchPractice) February 22, 2019
