search

muhanzhang / D-VAE

D-VAE: A Variational Autoencoder for Directed Acyclic Graphs, NeurIPS 2019
MIT License
127 stars 28 forks source link

Results in Bayesian Optimization #5

Open loukasilias opened 4 years ago

loukasilias commented 4 years ago

I have run the expirements with DVAE and in bayesian optimization i get the following results in best_arc_scores.txt : 118801345_672853366922853_1482046152836433382_n

Is this metric accuracy? Why is it nearly 10% ? Thank you in advance.

muhanzhang commented 4 years ago

Yes, the metric is accuracy. But the numbers shown are too low, which seems to be from an untrained model (so predicts each of the ten classes of CIFAR10 randomly). Did you download the pretrained ENAS model to "software/enas/" following README?

loukasilias commented 4 years ago

Thanks for your reply. You are right, i forgot to move the pretrained model to the remote environment. I see now that i get nearly 75% accuracy. Have you included any code in order to train fully the architectures as you mention in your paper, or have you just the weight sharing accuracy? Thanks in advance.

muhanzhang commented 4 years ago

Yes. You can copy your found architecture to fully_train_ENAS.py and run this file.