Thanks Ricardo for the quick reply. You’re right, I probably got too fixated on the cords and dims there, but the problem I really have is with the shape.
Thanks for the link to the Bambi multinomial model, I’m not sure if Bambi will be able to work long term for me, as I’ve integrated some masking into my pymc model (e.g. if a player plays 0 minutes, set to -inf) before the softmax, which I imagine I can’t do so easily in Bambi. But the pymc mode in the same link is also some food for thought!
I guess what I’m really after is suggestions to simplify the model above (instead of the separate arrays for each position category, use the original position array with all positions ), and declare a single “delta” with the dims being the number of different positions. I’ve tried this, but can’t get it to work so far… any thoughts about how I might go about this?