Support for alternative axis projection metrics in plots

Consider that emb is an embedding set and we run emb.plot(x_axis='man', y_axis='woman'). This would compute the normalized scalar projection (i.e. operator >) of every embedding in emb onto embeddings for 'man' and 'woman' so that they could be plotted in a 2D coordinate system.

Now the question is: why we should only stick to normalized scalar projection? For example, one might want to use another measure such as cosine similarity or euclidean distance to plot the embeddings.

• First of all, does this make sense? (to me, yes!)
• Secondly, is there currently an easy and built-in way to achieve this? (AFAIK, no!)
• Thirdly, is adding an argument to plot and plot_interactive methods to support this a good way? e.g. emb.plot(x_axis='man', y_axis='woman', axis_metric='cosine').

[mkaze]

I’m opening this issue because technically the issue should also be added to the plot_interactive methods.

@koaning Right! That’s why I mentioned “partially” in #192 😉

[koaning]

I’m opening this issue because technically the issue should also be added to the plot_interactive methods.