Improve ARAX's ranking algorithm

[chunyuma]

This is a super issue for "Better Ranking". I split it into the following few sub-tasks:

• [x] Adjust weights in the ranker. See the sub-task in #2300.
• [x] Add a function in ranker to give higher weight based on the number of edges. See the sub-task in https://github.com/RTXteam/RTX/issues/2301.
• [x] Gather examples and write a few test cases for evaluation. See the sub-task in https://github.com/RTXteam/RTX/issues/2302.
• [ ] Final evaluation after deployment.

[dkoslicki]

And from last week's AHM, here is a rough idea for specific ways to help improve the ranker:

Non-semmeddb direct “treats” should be ranked higher More edges is better Downweight text mining provider Downweight semmeddb Single drugbank edge better than multiple textmined edges

Gather examples in gold, silver, bronze, tin (handful, ~10) Set up test harness of these examples Load the results (no need to recompute) Re-rank Reassess Tweak, itterate

[chunyuma]

Keep this issue open until we have verified its effect in deployment.

[chunyuma]

Based on recent evaluation report from Eric, the new ARAX's ranking algorithm looks great. So, I close this issue.