Matrix Factorization Analysis

[patriacaelum]

We have the functions defined to use matrix factorization. There are a few things we should look into before expanding for all actions.

• How to normalize the output matrix of probabilities. Since it is a linear sum, it may be as simple as dividing by the number of features
• How to use those probabilities to decide on an action. Since there is a probability to fund every proposal, do we sample from a random proposal at each time step? Or do we roll on the most probable proposal? Or do we sample from each proportionally?
• Create a graph output that shows the probabilities are working as intended, and that the wallets that are funding are more likely to fund proposals within their interests

[theoriginalMAR]

1- I just updated the matrix factorization notebook, I used a normalization attempt from this site to normalize the matrix (https://www.mygreatlearning.com/blog/numpy-normalization-tutorial/#Normalization%20of%20Two%20Dimensional%20(2D)%20array). Let me know what you think
2- I think choosing the most probable proposal makes the most sense, or depending on the number of proposals available, we can pick the top 3 proposals to be funded. Now 3 is an arbitrary number, but I think it should scale to the amount of proposals present, it doesn't have to be fixed or anything.
3- We should probably discuss this one in our next sync up.

[xm3van]

2. Agree with Taz - I think set of most favoured proposal makes most sense because even if a funder has funds available the are unlikely to fund anything especially if they don't like the proposal very much. There is usually no expiry date for funds💵 - maybe it makes sense to define a threshold of how likely they fund a proposal and draw from proposals above the threshold (i.e. everything above 0.8 enters the pool of what can be potentially funded).