Person: I see you're training your neural net. How's it doing?
Me: Well...

lol..bruh..what net is that? architecture

Soooo, gain?

@bdhobare a lstm->fullyconnected->regression. Note that I wrote the loss function special. Nevertheless, you should not be able to get some negative when working with positive numbersand not a single negative ones, since you work with distances.

I must have revolutionized mathematics

@Kimmax sadly no, performed even worse that the previos one

I just hope this works faster than a month for your sake 😶

@Awlex use RELU before the regression layer to remove negatives

count your losses

I wouldn't mind some negative loss...

Doesn't look like it is converging