pender

I wrote a beam search implementation on top of char-rnn some time ago. See sample-beam.lua here: https://github.com/pender/char-rnn/blob/master/sample-beam.lua . A beam width of 2 on a fully trained model eliminates effectively...

I think it's recognized theoretically that widening the beam can result in an inferior result, and I seem to recall that the papers that mention beam search in the context...

> IMO it would make sense if discorded allies didn't allow you to switch places with them. > > Then, if you can't switch places with them, the only option...

Or maybe they should be able to spawn with an innate + or - enchant!

I would argue against free stacking of different sets of throwables, for what it's worth.

> Confirmed. This appears to be intentional. I did a quick test by removing the STATUS_STUCK check below in Movement.c, and the behavior was as expected. The monster successively got...

I vote to specifically force the cell containing the monster to be visible during the player's next turn as a special case, and have it show up in the sidebar...

> Visible as in completely visible, or greyed out like the other tiles you can't see? I'd say completely visible for all gameplay purposes. There's no light on the cell,...

I suspect this can be fixed by adding `(DFF_BLOCKED_BY_OTHER_LAYERS)` to the dungeon feature flags for the [Dryad grass DF](https://github.com/tmewett/BrogueCE/blob/master/src/brogue/Globals.c#L780), although it may prevent dryad grass from spawning on shallow water...

Yes, TPU would probably be much faster. If you really know what you're doing, you could try replacing my beam search implementation with one that happens entirely on the GPU,...