Anders Aasted Isaksen

M.D., PhD student, data enthusiast, diabetes and migrant health researcher

[Anders Aasted Isaksen @Aastedet]

@karpathy Something along the lines of RStudio’s support for markdown preview in the editor?: https://t.co/VKCo7acFBa

[ScienceNews @sciencenewsdk]

Migrants in #Denmark with type 2 #diabetes are monitored less than native Danes for damage to the kidneys, feet and eyes. They also had higher #cholesterol and blood glucose levels.

#sciencenewsdk @Aastedet

https://t.co/g4Rp6Taysd

[Peyman Milanfar @docmilanfar]

we thank the reviewers for helpful suggestions

[Anders Aasted Isaksen @Aastedet]

Just published the first paper of my PhD project in Clinical Epidemiology!
Found that register-based classifiers accurately identified type 1 and type 2 diabetes in a general population, but performed poorly in cases with atypical age at onset of diabetes. @StenoAarhusRes

[Diabetes Tweets @DiabetesTweets]

Validation of register-based diabetes classifiers | CLEP - Dove Medical Press https://t.co/2KSQSPdzWS

[Diabetes Tweets @DiabetesTweets]

Validation of register-based diabetes classifiers | CLEP - Dove Medical Press https://t.co/2KSQSPdzWS

[Adam Mastroianni @a_m_mastroianni]

Science ran a big experiment on itself for the last ~60 years. It didn't work out.
https://t.co/1pNOSRrdNY

[Anders Aasted Isaksen @Aastedet]

@DrLeo037 @vopani @tunguz @marktenenholtz That’s true (a shame, someone should adopt the benchmarking project and re-run the scripts), but the pydatatable benchmarks are still up to date, as it hasn’t had any update releases since then anyway.

[Anders Aasted Isaksen @Aastedet]

@vopani @tunguz @marktenenholtz It’s not nearly as fast as its R parent, and these days there’s not much speed difference between pydatatable and pandas for most operations. It still appears much more stable for very large datasets though: https://t.co/bghr4qkQOr

[Anders Aasted Isaksen @Aastedet]

@svpino You could argue extending the split even further to radiology unit, if you’re scared that differences between pictures from different x-ray machines (contrast, gamma, etc) might also leak info from training to validation set. Less impact than leaking individual-level data ofc.

[David Brückner @d_brueckner]

There's been lots of talk on moving #ScienceTwitter to another platform

As scientists, we should make informed decisions & not accept status quo - so what are the options? I checked out open-source non-profit @joinmastodon

How it works, why it's great & how to get started!🧵

[Anders Aasted Isaksen @Aastedet]

@marielli @Raspberry_Pi @arduino My first project was a variation of this face recognizer/party greeter: https://t.co/X3nLwTd3MT using PiCam and a BT speaker. Was a lot of fun!

[Anders Aasted Isaksen @Aastedet]

@abhi1thakur Or even better: cross-validate it with tons of data leakage to show stakeholders that sweet >99% accuracy. Whatever happens after that is Ops’ problem :)

[Anders Aasted Isaksen @Aastedet]

@fchollet Better than calling it a car-ensemble!

[Anders Aasted Isaksen @Aastedet]

@Xvayzet3 @marktenenholtz Semantics, except one raises funding rounds, the other pays the bills :)

[Anders Aasted Isaksen @Aastedet]

@badbit_0 @vopani Probably https://t.co/ANym7MtDNa
Originally an R package for fast/memory-efficient data-manipulation, now implemented in Python as well.

[Anders Aasted Isaksen @Aastedet]

@ScienceInsider Someday they might even realize that peer-reviewed, open-access journals without processing fees are completely feasible @JOSS_TheOJ

[Anders Aasted Isaksen @Aastedet]

@AnoopRKulkarni @GFaghe For open access datasets, there are none with 12-lead ECG+diabetes status, to my knowledge.
The closest is PTB-XL: 12-lead ECG data and BMI (might serve as a rough proxy of diabetes status): https://t.co/oIPpgGu2Nc
The open ECG+diabetes datasets only contain 1/2-lead ECGs, sadly.

[Anders Aasted Isaksen @Aastedet]

@AnoopRKulkarni @GFaghe ...
Or that these heartbeat features are simply specific of the adjacent heartbeats in the same ECG?
/end