This example trains a simple LSTM on a sequence-to-sequence addition task using an encoder-decoder architecture. The data is generated on the fly.
I0314 18:33:34.921972 139788256962368 train.py:280] train step: 9800, loss: 0.0004, accuracy: 100.00
I0314 18:33:35.791534 139788256962368 train.py:249] DECODE: 25+45 = 70 (CORRECT)
I0314 18:33:35.791721 139788256962368 train.py:249] DECODE: 27+92 = 119 (CORRECT)
I0314 18:33:35.791795 139788256962368 train.py:249] DECODE: 51+420 = 471 (CORRECT)
I0314 18:33:35.791843 139788256962368 train.py:249] DECODE: 49+450 = 499 (CORRECT)
I0314 18:33:35.791887 139788256962368 train.py:249] DECODE: 48+853 = 901 (CORRECT)
python train.py