More memory doesn't help OOM errors for #sklearn. That small bug in feature engineering used like 1 billion TB of RAM. Hafta fix the bug.
#SKlearn
You read that right, #SKlearn has a passive-aggressive model. http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.PassiveAggressiveClassifier.html It is a total asshat too. https://x.com/randal_olson/status/898538252608700416
I've got ideas for using ip-to city, ssn deceased file & a name popularity file to add features to a dataset for #sklearn consumption.
META: Why #sklearn thinks people follow me back & what I did with that info : https://medium.com/@mistersql/why-people-follow-back-according-to-sklearn-f8607ff55aad
I think in this case-machine learning just restated what in retrospect was obvious & said it emphatically enough for me to take action on it
♡ 1 ↻ 0
#sklearn Machine learning Q: Predictors with no theoretical reason why they should have an impact- leave them in or manually take them out?
#Sklearn is finally making reasonable predictions. What helped:
- more data
- throw away apriori garbage data
- viewing feature_importances_
Me trying to get #sklearn to make predictions better than a DummyClassifier https://x.com/inechii/status/892838683560341504
Okay, #Sklearn models are modeling (i.e. not throwing errors and blowing up), but they don't outperform the DummyClassifier. <Sigh>
If the office rules say no portable electric heaters, does that prohibit running #SkLearn on large datasets on my mac?
Is there any 3-5 day #scikitlearn training happening in DC over the next year? #sklearn Or anywhere for that matter?