search

tigvarts / vaeac

Variational Autoencoder with Arbitrary Conditioning
MIT License
79 stars 23 forks source link

Q: Giving full information to proposal network? #5

Closed geonyeong-park closed 4 years ago

geonyeong-park commented 4 years ago

https://github.com/tigvarts/vaeac/blob/80110ba55ed6b6362ddeea988980c007bb4f23f9/VAEAC.py#L57-L71

Hello, thank you for your great repo, and i have a question in the inferencing part: Is it valid to give the full batch without explicit masking into proposal network? Since prior network, which will be used in test phase, is provided masked batch as an input.

geonyeong-park commented 4 years ago

sorry it was my misunderstanding. Training loss from observed region is neglected during optimization.