I'm using Talos to hyper tune a simple convolutional network.
After running the scan function the performance looks impressive (val loss & val acc). When using the same params to train my network again it doesn't get even close. Feels like the params and the performance have nothing to do with each other.
I am getting the params as follows: r.best_params()[0] Where r is the reporting object.
How can I reproduce the same performance?
Asked
Active
Viewed 74 times
1
idoshamun
- 1,125
- 9
- 21