enas
enas copied to clipboard
Question about new searched rnn cell....
Thank you for open your work. Now I tried some experiments. I tried the search part for rnn cell, then searched rnn cell retrained. However, I can't reach the your paper result. It is very low performance. My result is above.. cell : 0 0 0 0 2 0 0 0 0 0 0 3 2 4 0 0 1 7 0 5 0 10 0 -> test ppl : 64.71 cell : 0 0 3 0 1 0 3 0 3 0 1 0 3 0 3 7 0 6 0 9 0 10 0 -> test ppl : 69.66 i checked your fixed architecture. it reaches the 55.8 ppl. cell : 0 0 0 1 1 2 1 2 0 2 0 5 1 1 0 6 1 8 1 8 1 8 1 -> test ppl : 55.8 ppl Why is different result between the your searched cell and my searched cell ?? In final training, How adjust the hyper-parameters??