Not Achieving Same Accuracy Results

[ne1410is-mc]

Hello, I'm running this code from my local machine. It's a direct fork from this repo. The only updates I've made are to make it compatible with Python3. Are you certain you're seeing ~95% accuracy when running with these hyperparams?:
--classes 5 --shots 5 --train-shots 10 --meta-iterations 100000 --iterations 5 --test-iterations 50 --batch 10 --meta-lr 0.2 --lr 0.001

Because I am not. Here is my final log output:
Meta-train
average metaloss 0.9020509621277452
average accuracy 0.7628000104278326

Meta-val
average metaloss 0.9079857150167226
average accuracy 0.7575000119954347

Any ideas why I may be seeing much lower accuracies? Any input would be appreciated. Thanks.

[gabrielhuang]

Sorry, it's been a while. Ideally try to run on Python2, there are a few subtleties that might have broken during conversion. You might also want to try lower meta-lr/lr

…

On Wed., Mar. 4, 2020, 10:55 a.m. Maria Cioffi ***@***.***> wrote: Hello, I'm running this code from my local machine. It's a direct fork from this repo. The only updates I've made are to make it compatible with Python3. Are you certain you're seeing ~95% accuracy when running with these hyperparams?: --classes 5 --shots 5 --train-shots 10 --meta-iterations 100000 --iterations 5 --test-iterations 50 --batch 10 --meta-lr 0.2 --lr 0.001 Because I am not. Here is my final log output: Meta-train average metaloss 0.9020509621277452 average accuracy 0.7628000104278326 Meta-val average metaloss 0.9079857150167226 average accuracy 0.7575000119954347 Any ideas why I may be seeing much lower accuracies? Any input would be appreciated. Thanks. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#7?email_source=notifications&email_token=AB3P5RBHK3N4YCJYINSVHCTRFZ2XNA5CNFSM4LBLSO7KYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4ISN3WTA>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3P5RANPZ35OJHN2MC4ANDRFZ2XNANCNFSM4LBLSO7A> .