Milestone missed; model will be terminated. #22
Description
Hi!, This neuron network is awesome! But when I tried to train the model by myself using CSAP7, I encounter a problem named "Milestone missed; model will be terminated."
This problem was raised because my training process did not reach the milestone defined at config file. The validationMilestone field in CASP7's configuration is {1000:14,5000:13.6}. However, I've tried it on several different machines, the loss of my training process at 1000 step can only down to 16.907.
I know the problem could be solved if I adjust the validationMilestone field in the configuration file, but I am wondering why I can never reach the milestone set by default. Is this due to the performance of GPU or other tricks?