Updated version? Issues + performance about python-glad HOT 3 CLOSED

Hi, thanks for contacting. This implementation is just for learning how GLAD works. So I focused on the readability rather than the stability and efficiency. (You could improve it by using numpy matrix computation more.) If you're interested in using GLAD for your work, you might want to try Whitehill's C Implementation available from his website.

from python-glad.

Oh, cool, thanx, this might be very useful! The only thing that this implementation lacks is multilabel mode. But in my case it's not true multilabel but some of "grey" between "black" and "white". I may find a right way to hack binary algorithm for the trinary judgements.

Another point of improvement that can be used for production is a kind of "incremental" fast mode. When we add some new judgements and some new workers. For this case a single iteration that starts with backed up alpha and beta distribution could be sufficient. But also this one could be too long, so perhaps it's worth to implement partial iteration when you skip workers and labels that were not affected by the new data.

from python-glad.

Thanks for your suggestion for speeding up!

There are some algorithms proposed for ordinal labels like white->gray->black. For example, see Zhou et al., Aggregating Ordinal Labels from Crowds by Minimax Conditional Entropy. ICML2014. The implementation can be found at the author's website.

from python-glad.