These held out examples can become your training data.
Now your filter is blocking at least 74% of the negative examples. You can gather cleaner data if instead during serving you label 1% of all traffic as "held out", and send all held out examples to the user. > But this approach introduces sampling bias. For example, if a user marks an email as spam that your filter let through, you might want to learn from that. You might be tempted to draw additional training data from the instances shown to users. Suppose you have a filter that blocks 75% of the negative examples at serving. > In a filtering task, examples which are marked as negative are not shown to the user. > Rule #34: In binary classification for filtering (such as spam detection or determining interesting emails), make small short-term sacrifices in performance for very clean data. See this quote from Google's "Rules of Machine Learning" (A great resource by the way) Google probably lets some amount of known-spam emails through for data gathering. It works but kind of annoying going through setting up each domain added through the relay. I've been using Mailu via docker-compose on DigitalOcean for a couple years for all my lesser used domains/addresses, relaying through SendGrid. But 10x-ing my costs just doesn't feel right for something that will take more time and not generate revenue that I'm not that passionate about.įor those curious, been looking at WildDuck mail which seems like an interesting structure and the features are cool, just not sure I want to go through it all. I've been considering setting up a higher end server (compared to the $20/mo vps I'd been using) at a data center and seeing what I can manage as a direct mail host without the relay. I'm relaying through SendGrid as I just don't have that many emails coming from/through my server that it's worth the lowest paid level (there is a free tier) to have to worry about it. I can get through to hotmail and o365, but not the block. And good luck getting off that list if you're on a hosted VPS.
0 kommentar(er)
