implemented a umap trajectory feature

[danich1]

This PR implements the backed to propagate umap coordinates onto the frontend to provide users an alternative way to see how neighborhoods are shifting. I received feedback that this feature may help some audience understand the word neighbor changes. My code works locally, but there may be an issue when its fully implemented onto the online server.

In regards to the front-end, I'm envisioning a simple scatter plot each point is a token and they have their corresponding labels next to them. The query token would have a new shape or some indication that this was the token they queried and lastly, an arrow to show the trajectory path (tok_2000 -> tok_2001...-> tok_2021).

The returned object is now in the form of:

  {
        "neighbors": {<year:str>: [<token:str>, ...]},
        "umap_coords": [{"year":<year:int>, "is_query":<is_query:bool>, "token":<token:str> ,"umap_x_coord":<umap_x_coord:float>, "umap_y_coord":<umap_y_coord:float>}, ...],
        "frequency": [{"year": <year:int>, "frequency": <frequency:float>}, ...],
        "changepoints": [ [<year_start:str>, <year_end:str>], ...],
        "elapsed": <elapsed_ms:float>
    }

[netlify[bot]]

✅ Deploy Preview for word-lapse canceled.

Name	Link
🔨 Latest commit	0d7674ca7420302dc24528620e485b5d71b703df
🔍 Latest deploy log	https://app.netlify.com/sites/word-lapse/deploys/6262e9cc26041200088c962f

[vincerubinetti]

Nice, could you post some sample json just for the umap. I have time today to start working on the visualization.

[danich1]

Here is an example demo json. The tokens aren't de-identified; however, they will be in the actual implementation. umap_demo.json.txt