David Benkeser

1.2K posts

David Benkeser banner
David Benkeser

David Benkeser

@biosbenk

Assist. Prof. Biostat @EmoryRollins. he/him. machine learning. causal inference. vaccines. data science. all views mine. all math done bored on conference call.

ATL, GA Katılım Ağustos 2020
246 Takip Edilen1.7K Takipçiler
David Benkeser
David Benkeser@biosbenk·
Is there a word in German for “having a paper rejected and a positive Covid test within 1 minute of one another”? Kränklichetraurigkeitablehnung?
English
6
0
18
0
Daniela Witten
Daniela Witten@daniela_witten·
Asymptotically, my jokes are a 10. Just wait for the sample size to increase a bit
English
8
12
285
0
David Benkeser
David Benkeser@biosbenk·
@MaartenvSmeden Caveat that bootstrapping is not theoretically justified for all learning algorithms and can struggle with the heavy fitters (Eg random forest). Fine for simpler algorithms like vanilla logistic regression. CV provides the most general solution.
English
1
0
2
0
Maarten van Smeden
Maarten van Smeden@MaartenvSmeden·
Should also have mentioned bootstrapping here...
English
3
0
7
0
Maarten van Smeden
Maarten van Smeden@MaartenvSmeden·
Good description of leakage, but very often an even better solution is not to split your data into train-test sets at all. Cross validation and internal-external validation should be the starting point (deviate only when necessary)
Santiago@svpino

Here is what Machine Learning tutorials told you to do: 1. Start by transforming your dataset 2. Then split it (train, validation, and test sets) 3. Finally, build your model Please, unlearn this process. There's a problem with it: 1 of 11

English
6
17
113
0
David Benkeser
David Benkeser@biosbenk·
@UnibusPluram Studies show that 7/8 is the preferred time signature for infants under 6 months old.
English
0
0
1
0
Alejandro Schuler
Alejandro Schuler@UnibusPluram·
@biosbenk Yesterday my daughter thoroughly enjoyed listening to The Mars Volta during tummy time. Pretty sure the head bopping was her digging the odd time signatures and had nothing to do with her neck being barely strong enough to hold her head up.
English
1
0
1
0
David Benkeser
David Benkeser@biosbenk·
In case anyone was wondering, making stuffed animals headbang to Metallica is in fact high entertainment to 3 month olds.
English
2
0
7
0
David Benkeser
David Benkeser@biosbenk·
Follow me during paternity leave for more terrible, sleep-deprived stats/parenting jokes!
English
2
0
8
0
David Benkeser
David Benkeser@biosbenk·
Parents are all out here trying to lengthen that right tail… more kurtosis please!
English
1
0
2
0
David Benkeser
David Benkeser@biosbenk·
Next time I need uniform random numbers generated, I will just bootstrap the distribution of nap lengths from my three month old.
English
3
0
10
0
David Benkeser
David Benkeser@biosbenk·
@NickytaLeb I’m good with that progression. If my kids first words aren’t the lyrics to Hangar 18, have I really done anything as a parent?
English
0
0
2
0
David Benkeser
David Benkeser@biosbenk·
@jon_y_huang You’re supposed to leave one observation out of a cluster Jon. That’s what leave one out cross-validation is. - some guy on Stack Overflow, probably
English
1
0
3
0
David Benkeser
David Benkeser@biosbenk·
@StatsSimon I have been wondering about this problem for a while! Very cool result. Congrats!
English
0
0
1
0
Noah Simon
Noah Simon@StatsSimon·
What does this say about deep reinforcement learning? Perhaps we can consider starting with a simple network and scheduling additions to the topology as we go --- though possible that we want to decrease step size for those additions. (8/n) Sorry this thread was so long!
English
1
0
2
0
Noah Simon
Noah Simon@StatsSimon·
Wanted to share a lovely paper primarily by @StatTZhang that was just accepted at Annals of Statistics on efficient non-parametric regression using stochastic optimization! arxiv.org/abs/2104.00846 Less technical summary to follow... (1/n)
English
1
2
16
0
David Benkeser
David Benkeser@biosbenk·
@DrJWolfson Tik Tok of you doing the rap verse or it didn’t happen. Stackity stackity…
English
0
0
0
0
David Benkeser
David Benkeser@biosbenk·
Just taught a class using the RStudio IDE for the first time in years and…still don’t love it. I know I may be a dying breed but like omg let me choose my own file extension.
English
1
0
7
0
David Benkeser
David Benkeser@biosbenk·
@alexpghayes @statsepi I mean that without more assumptions there are a multitude of curves that are equally likely given those data. The data will not be able to tell you which is true.
English
1
0
0
0

Keşfet

@edwardhkennedy @daniela_witten @MaartenvSmeden @UnibusPluram @NickytaLeb @StatsSimon @elonmusk @BarackObama