Minimize disk usage

[simonlindholm]

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]

Everything is in the cloud nowadays! :p

[simonlindholm]

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

[HalfVoxel]

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

[simonlindholm]

Yes. This is for later.

[HalfVoxel]

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).