I don’t understand why does my embeding projector chart is so different than one which was shown in the video? I understand that when generating models, it can produce different weights, but they are not even close to be similar… is it ok?
Video:
As long as you ran the notebook and downloaded the vecs and meta tsv files, I wouldn’t worry too much about the overall structure. Here’s the visualization on my run:
The point to pay attention is that words that are similar will appear closer than words that are dissimilar. For instance, words like worst and awful will be a lot closer than worst and fantastic. Please see the images below:
