Minimize disk usage - Githubissues

Gullesnuffs / Codenames

A codenames bot playing the part of the spymaster.

19 stars 8 forks source link

Minimize disk usage #10

Open simonlindholm opened 7 years ago

simonlindholm commented 7 years ago

So it fits on a phone, for instance. Currently the word2vec model uses ~50MB, which we could reduce by:

using fewer words, winning say a factor of 2
using half-floats instead of floats, winning a factor of 2
storing similarities between each dictionary word and Codenames word instead of vectors, winning a factor of 300/100=3 (makes it impossible to play with custom words, though)

HalfVoxel commented 7 years ago

Everything is in the cloud nowadays! :p

simonlindholm commented 7 years ago

Eh, local stuff is nicer if we can get away with it. Less code, lower latencies and offline usage.

HalfVoxel commented 7 years ago

In any case I think the focus should be on improving the accuracy of it instead of the memory usage.

simonlindholm commented 7 years ago

Yes. This is for later.

HalfVoxel commented 7 years ago

using half-floats instead of floats, winning a factor of 2

Do we even need half-floats? Using fixed point numbers with 1 byte per component should be enough (assuming normalized vectors the range is between -1 and 1 so the resolution is 2/255 ≈ 0.008).