I'm wondering if there is any reason you have used "model-100.net" as encoder initialization when you train decoder, this line?
When I use "model-best.net" as pre-trained encoder, my decoder best number is as follows?
Best test error: 0.75259215056896, in epoch: 79
Is it similar to what you get?
I'm wondering if there is any reason you have used "model-100.net" as encoder initialization when you train decoder, this line? When I use "model-best.net" as pre-trained encoder, my decoder best number is as follows? Best test error: 0.75259215056896, in epoch: 79 Is it similar to what you get?