Aaron (1e6c3b88) at 01 Aug 15:04
Karel van de Plassche (85bb6989) at 01 Aug 15:04
Merge 'Active learning NN, collaboration with UKAEA' into 'main'
... and 15 more commits
evaluate_network
to process nets with multiple outputs, but still only returns 1 output (first output of given NN)discretize_classifier_output
converts real number output to "indices" based on input thresholdsimpose_classifier_constraint
multiples classifier output column with other columns based on mapadept
data/README.md
to include new repository for adept
weights and biases (currently a private repo)evaluate_network
to process nets with multiple outputs, but still only returns 1 output (first output of given NN)discretize_classifier_output
converts real number output to "indices" based on input thresholdsimpose_classifier_constraint
multiples classifier output column with other columns based on mapadept
data/README.md
to include new repository for adept
weights and biases (currently a private repo)Good idea, I made the change.
Yes, I did test the standalone version. It seems to work fine. However, since this is likely not the final version of the weights and biases, I do not want to put a CI test in just yet. I will add the net name to namelist_tools.py
.
Ah maybe I am misinterpreting this comment. Do you just want me to remove the commented out line rather than remove the change?
It is necessary to remove this stop, since the new ADEPT networks have n_outputs = 2
. I have modified evaluate_network()
to only return the first output, making the function consistent with the previous implementation.
Then, I also removed the variable error_msg
itself since this is the only usage of it in the function.
Some small comments. Did you test the standalone version? Maybe also add your net to tools/namelist_tools.py to generate a reference input namelist to test with.
where (a < 0)
y = 0
elsewhere
y = a
endwhere
Might be slightly more efficient
Remove
Remove
Remove
The network architecture and code infrastructure is more or less settled, so the merge can happen. The weights and biases data files may still change, but this can be done in another MR.
Aaron (cec2b7d0) at 30 Jul 22:14
Adjusted Makefiles to reflect ALNN name change to ADEPT
Aaron (fb070bcc) at 30 Jul 22:06
Added namelist data files for QLKNN-ADEPT, preliminary version
Aaron (fac6fd1c) at 30 Jul 21:47
Aaron (b1698a91) at 30 Jul 20:45
Added namelist data files for QLKNN-ADEPT network
... and 1 more commit
The current version of RAPTOR refers to commit c36e3318. This commit exists on the remote but is not on a branch. Some users are having trouble checking out this commit.
I suggest creating a branch at that location called RAPTOR-is-here
so that git fetch
will fetch
the remote version.