bnelearn.mechanism.auctions_single_player module

class bnelearn.mechanism.auctions_single_player.StaticFunctionMechanism(cuda: bool = True)

Bases: Mechanism

A static mechanism that can be used for testing purposes, in order to test functionality/efficiency of optimizers without introducing additional stochasticity from multi-player learning dynamics. This function more straightforward than the Static Mechanism above, which has stochasticity similar to an auction.

Instead, this class returns a straight up function, designed such that vanilla PG will also work on it.

Here, the player gets the item with probability 0.5 and pays (5-b)², i.e. it’s optimal to always bid 5. The expected utility in optimal strategy is thus 2.5.

run(bids)

Alias for play for auction mechanisms

class bnelearn.mechanism.auctions_single_player.StaticMechanism(cuda: bool = True)

Bases: Mechanism

A static mechanism that can be used for testing purposes, in order to test functionality/efficiency of optimizers without introducing additional stochasticity from multi-player learning dynamics.

In this ‘single-player single-item’ setting, items are allocated with probability bid/10, payments are always given by b²/20, even when the item is not allocated. The expected payoff from this mechanism is thus b/10 * v - 0.05b², The optimal strategy fo an agent with quasilinear utility is given by bidding truthfully.

run(bids: Tensor)

Alias for play for auction mechanisms