search

CW-Huang / torchkit

MIT License
37 stars 9 forks source link

Different use of word 'context' in code compared to paper? #4

Closed Binbose closed 5 years ago

Binbose commented 5 years ago

Hey, awesome project! However, do I understand it correctly, that you use the word 'context' differently in your code, compared to your paper https://arxiv.org/pdf/1804.00779.pdf In the paper, context is a function depending on the previous dimensions, whereas in the code, it seems like context is a variable, that is set constant to 2.0 (I assume, it can be used to condition the network on some additional input, is that correct?)

CW-Huang commented 5 years ago

you're right: context in the code is used for conditioning.

In the paper we also use the word "context" to describe conditioning on data in amortized inference (VAE), see Appendix H. Here's a piece of description quoted from there: "... The convolution layers are followed by a fully connected layer of size 450 as a context for the flow layers that transform the noise sampled from a standard normal distribution of dimension 32. ..."

Binbose commented 5 years ago

Ah sorry, I mixed it up with conditioner, I should have been more careful. Thank you though!