The following program is an implementation of making an application that creates the optimal word for your team in Word 2 Vec.
Guidance from this article and this medium post. Using the GoogleNews-vectors-negative300.bin dataset at the start, as inspired by Semantle. It's a bit funky though, so I may look into other datasets out there, especially with better stop words.
The documentation on gensim is also very guiding, but mostly I just poke around the source since the documentation is generated from that.
My laptop currently uses 64-bit Python, has 32gb of RAM, and it takes about 48 seconds to load the model the first time- but it saves the gensim model after that. I don't think there's anything else particular relevant for that? I don't think my console can use more than a certain allocation of RAM though.
More or less, this is out of the box. You'll want to pip install the following dependencies:
- nltk
- node2vec
- gensim
- numpy
- May need to do
pip install numpy --upgradeif you get an error with gensim
- May need to do
For the simulation and analysis stuff:
- matplotlib
- scikit-learn
- pandas
Create a folder called model as that is where I save gensim models and where they (probably) should be loaded from.
And then you'll need Google's dataset, which- because it's 1.5gb- is not included in this repository. Download from Google Drive here. You can also use gensim's downloader module to download various datasets and the like, there's a whole bunch of experimentation I guess. I just don't want to train my own model.
Just a way to briefly expose myself to finally implementing this kind of thing.
In its most basic form, you input all 25 words on the codenames board. Then you give it the 5 words you need ot make a hint for. And then it spits out a one word (potentially two word? idk) hint given the most related words.
TODO: Type up that sheet of paper I have my notes on and slap it into here
To "have fun with it", maybe add in some graphs or analyses or something as well?