jdgh000
commented
1 month ago Alright, printing full last two corners got following, 3rd corner also +, - signifying random nature. WIll be interesting to read the output once trained. Can close this issue or open a while is fine just to bring up interesting points.
['hidden_seq'] : <class 'torch.Tensor'>
torch.Size([1, 2, 2])
tensor([[[ 0.0832, -0.0356],
[ 0.3105, -0.5263]]], grad_fn=
I totally got it that code has no training at all in starting example of enc-dec example around pages ch9 p227-p233 where two corners were fed into enc-dec model and final corner of output of two corner is displayed. I see example input is starting from lower left going CCW. Final corner in the example is obviously [1,-1]. But the predicted final output is I am also seeing tensor([[[ 0.3105, -0.5263]]], grad_fn=)
while it is not close to 1,-1 but it at least got the coordinate panes right (+, -), wondering how did it so, by pure luck??