I use deep learning to try identify which pictures I’ve taken are good (using a loosely defined version of ‘good’). The resultant convolutional neural network (CNN) performs well (0.81 AUC), helped me identify some of my own biases, and seems like a fairly interesting start for more analysis.
I don’t know whether to feel happy or sad that scammers are A/B testing poorly. I recently received two emails, that were obviously part of an experiment done poorly. They’re shown below, but a few notes for a potential future scammer.
For the past few months, the press has been focused fixated on whether Uber will be forced to reclassify drivers from contractors to employees, it has ignored a bigger question: what would be the effect of Uber employing millions of drivers?
I’ve noticed a frustrating phenomenon: the use of linear scales where a log scale would be much more appropriate. While people are more conscientious of deceptive charts now more than ever, misleading scales continue to be an issue. This is my brief rant against them.