5 ms·

It's very much not trivial, though. Given two solutions to a coding problem, how do you objectively determine which one is better? More importantly, how do you

by s73ver 9y ago

It's very much not trivial, though. Given two solutions to a coding problem, how do you objectively determine which one is better? More importantly, how do you objectively tell if either of them is good enough? Or if they're poor?

bitL 9y ago
Probably wrong granularity; we would likely have to wait a bit for this detailed ML. But you can make various groups that follow certain "ideology" and see which ones thrive under what circumstances/in what kind of environment or not. Kinda like A/B testing but internally on their own coders. Then you can view e.g. maintenance costs (that might be your code quality metrics aggregated over time) etc. If there is one company that can pull this off, it's Google. I would be very interested in results to be honest. Then this could be generalized for governments and provide fact-based guidelines on how to govern well.
- s73ver 9y ago
  Unfortunately, how do you control that both groups are the same except for the metric you're trying to measure? And, once again, how do you accurately measure maintenance costs, especially over different projects that would likely have variations in requirements? And, most importantly, how do you come up with objective measurements for these things?
  - bitL 9y ago
    First, it's going to be an approximation and would take decades to carry out, as we humans are slow and no GPU/cloud would accelerate us :-( Second, the idea I had on mind was to use ML to figure out which signals/features ire the most important ones; just dump all raw data onto it and expect it would figure it out. Kinda like what you see from PCA which gives you base vectors guaranteeing some correlation properties, or CNNs where you expects neural network to come up with convolutions it needs on its own. With maintenance costs I'd probably record as much data as I can and then try to find in a given case what kinds of signals were present. If I see some are repeating and are beneficiary or not, I can have some classifier from them. Also, it could unearth parasitistic relations when certain group of individuals flourishes only because they take advantage of certain individuals with different ethics, but never thrive in other kinds of environments.

by s73ver 9y ago

bitL 9y ago

Probably wrong granularity; we would likely have to wait a bit for this detailed ML. But you can make various groups that follow certain "ideology" and see which ones thrive under what circumstances/in what kind of environment or not. Kinda like A/B testing but internally on their own coders. Then you can view e.g. maintenance costs (that might be your code quality metrics aggregated over time) etc. If there is one company that can pull this off, it's Google. I would be very interested in results to be honest. Then this could be generalized for governments and provide fact-based guidelines on how to govern well.

s73ver 9y ago
Unfortunately, how do you control that both groups are the same except for the metric you're trying to measure? And, once again, how do you accurately measure maintenance costs, especially over different projects that would likely have variations in requirements? And, most importantly, how do you come up with objective measurements for these things?
- bitL 9y ago
  First, it's going to be an approximation and would take decades to carry out, as we humans are slow and no GPU/cloud would accelerate us :-( Second, the idea I had on mind was to use ML to figure out which signals/features ire the most important ones; just dump all raw data onto it and expect it would figure it out. Kinda like what you see from PCA which gives you base vectors guaranteeing some correlation properties, or CNNs where you expects neural network to come up with convolutions it needs on its own. With maintenance costs I'd probably record as much data as I can and then try to find in a given case what kinds of signals were present. If I see some are repeating and are beneficiary or not, I can have some classifier from them. Also, it could unearth parasitistic relations when certain group of individuals flourishes only because they take advantage of certain individuals with different ethics, but never thrive in other kinds of environments.