Visualising Embedding Layer

[gauravreddy08]

I know I'm late to the show but, I was watching the Lecture 236, and here's something I noticed.

@mrdbourke might have pointed this later in the course but, as I said, my knowledge is limited to the end of Lecture 236 only.

So when Daniel tries to visualise the embedding layer on the tensorflow projector website, here'a something I did.

I essentially saved weights of the embedding layer, before and after training it on the data. And visualised the both on the website, and here's what I found,

Before training:

After training:

If you're like me, it's interesting to see how the model updated the weights. In the first image, the points where all over the place, where as in the later one, the model started to figure out to classify between Disaster and Not Disaster.