What steps will reproduce the problem?
1. Train and save a model
2. For a new executable, load the variables/data in a different order
3. Load the previous model and evaluate
What is the expected output? What do you see instead?
The accuracy should be unaffected. However, since the model loading does not
check whether the indexing in file is same as indexing in Domain, weights could
"mean" different things.
Original issue reported on code.google.com by sameeersingh on 3 Mar 2012 at 12:43
Original issue reported on code.google.com by
sameeersingh
on 3 Mar 2012 at 12:43