Blogging Blitz 2020

The end of the year is just around the corner once again, and once again I have generated just a couple of actual blog posts, several times as many unfinished drafts, and dozens of hastily scribbled ideas. Again this year I'm taking some time off and am going to make a push to get as many posts out as I can.

With last years "blogging blitz" I managed to get out 2 extra posts before the year was out. Naively I had thought I might be able to get 10 or 12 "quick" posts finished. My working theory was that since I had more than half a dozen very detailed partially finished blog posts it should have been possible to extract small relatively complete portions of those posts and polish them up for quick publication. That along with a handful of quick explorations of completely new ideas could get me all the way to my (apparently rediculously over ambitious) goal of 1 post a day average.

But what actually happened was rather different. Each time I revisited one of unfinished blog posts into which I have already poured a significant amount of work there would be one or more compelling but difficult to solve problems laid out for me. In short this strategy amounted to one where I was continually nerd sniping myself.

Of the posts that I actually managed to put out in last years blogging blitz one of them was a quick treatment of a fresh idea (the permutation-cnn post) and the other was a somewhat more detailed write up of the application of the exponential moving average trick to the calculation of streaming Fourier transforms. The streaming Fourier transform started out as an extraction of an idea from a much longer half finished post (which is still languishing for want of attention). However almost none of the code or prose came from the parent post I think trying to incorporate it actually may have slowed me down.

This year I am planning on a very different strategy. Instead of trying to focus on ideas into which I have sunk the most time in the past, I am going to actively try to avoid things which have already received a lot of attention. Instead I am going to focus mostly on ideas that I have put little or no time into previously. I will also try and somewhat favor the simpler more well defined ideas over the more complicated or unusual ideas (though how fun I think the idea is will definitely also still factor in).

A lot of my half finished posts are of the form "achieve the ambitious but somewhat vague goal x". For example I have put a hundred hours or more into a post about attempting to turn the path through parameter space of a neural network as it trains into audio in a way such that you could use it as a training diagnostic tool. Apparently that is a difficult task (or at least very difficult for me). Incidentally, that is the post that I extracted the idea for the "streaming spectral summaries" post in last years blogging blitz. But for this years blitz I will do my best to keep from doing any such goal based posts (as opposed to idea/concept based).

Additionally I will try and intersperse the longer more involved idea driven posts with short "tips and tricks" type posts which focus on simple tricks which I find useful (for example perhaps a post about why you maybe should sometimes be using np.newaxis instead of the .reshape methods of numpy arrays).

The blogging blitz 2020 starts now!