Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Recurrent models should accept nonzero initial state #81

Open
lmjohns3 opened this issue Jun 24, 2015 · 1 comment
Open

Recurrent models should accept nonzero initial state #81

lmjohns3 opened this issue Jun 24, 2015 · 1 comment
Labels
feature
Milestone
v0.8.0

Comments

@lmjohns3
Copy link
Owner

To sample sequentially from a recurrent model, it would be useful to add some API for accepting an initial hidden state and running the model forward one time step.
@lmjohns3 lmjohns3 mentioned this issue Nov 29, 2015
Closed
@lmjohns3 lmjohns3 added this to the v0.8.0 milestone Dec 3, 2015
@x75
Copy link

x75 commented Feb 16, 2016

👍 , really looking forward

lmjohns3 pushed a commit that referenced this issue Feb 20, 2016
Allow symbolic initial recurrent state.
d90eb62 
Traditionally the initial state for a recurrent network is set to all
zeros, the assumption being that recurrent models will run for
sufficiently many time steps to amortize the effect of the initial
state. This change permits the initial state vectors to come from a
network output instead, which should allow for single-time-step model
stepping.

Contributes to issue #81.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
feature
Projects
None yet
Milestone
v0.8.0
Development

No branches or pull requests
2 participants
@lmjohns3 @x75