You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I’m planning to conduct research on multi-slot auction examples and am currently checking which parts of the code need modification.
Below is what I have identified so far and need your validation.
The classes in AuctionAllocation.py seem to already implement the logic for returning winners and prices up to the nth position.
🔧 since bid = value * eCTR the ranking is based on eCPM.
to implement GSP in this context we need to add eCTR to allocation input,
and set $price_i = bid_{i+1} / eCTR_{i}$ ?
The click allocation part in Auction.py also appears to support multiple slots.
🔧 For multi-slot auctions, we want to consider position effects.
to implement this in the click allocation we can do like $outcomes = binomial(1, CTRs * posEff)$ ?
If there are any additional areas in the project’s code that require modification for conducting research on multi-slot auctions,
I would greatly appreciate your guidance.
If it is structurally impossible for the project to support multi-slot auctions, please let me know.
The text was updated successfully, but these errors were encountered:
auction-gym/src/main.py
Lines 35 to 37 in 065f8bf
I’m planning to conduct research on multi-slot auction examples and am currently checking which parts of the code need modification.
Below is what I have identified so far and need your validation.
auction-gym/src/AuctionAllocation.py
Lines 26 to 35 in 065f8bf
to implement GSP in this context we need to add eCTR to allocation input,
and set
auction-gym/src/Auction.py
Line 65 in 065f8bf
to implement this in the click allocation we can do like
If there are any additional areas in the project’s code that require modification for conducting research on multi-slot auctions,
I would greatly appreciate your guidance.
The text was updated successfully, but these errors were encountered: