Now my learning process uses 90% of the GPU. However, the loss values increase. Did I just get a fast program that's also wrong? 4 atbildes
social.3dots.lv/@pixel (2017-01-13 18:15:55) |
Now my learning process uses 90% of the GPU. However, the loss values increase. Did I just get a fast program that's also wrong? | ||
social.3dots.lv/@pixel (2017-01-13 19:31:28) |
Orange: what it used to be. Blue: what it is now... https://t.co/4NA4fkwvML | ||
social.3dots.lv/@pixel (2017-01-13 19:39:23) |
I've also changed the optimizer: gradient descent to Adam. Now the efficiency and effectiveness are back. https://t.co/S44XRBTxrm | ||
social.3dots.lv/@pixel (2017-01-13 19:40:20) |
The morale of the story is to change one thing at a time. | ||
social.3dots.lv/@pixel (2017-01-13 19:47:40) |
Another lesson is to read the docs. Read them carefully!!! https://t.co/NRmx0XyHzU |