Modify System class and add Session class

[anushabala]

• Add new_session() function (and any other required functions) to System
• System should load up the model at startup
• Session should provide an interface to send/receive messages from the model

[hhexiy]

Hey I just pushed some changes to neural_system and neural_session to the system branch. There are details to figure out but the interface is completed. Briefly, we load the TF model and launch a TF session in system; then in each agent session we will execute computation nodes in the loaded TF graph.

[anushabala]

Awesome, thank you!! :) So the web interface will just have to call send() and receive(), right?

On Wednesday, September 28, 2016, hhexiy notifications@github.com wrote:

Hey I just pushed some changes to neural_system and neural_session to the system branch. There are details to figure out but the interface is completed. Briefly, we load the TF model and launch a TF session in system; then in each agent session we will execute computation nodes in the loaded TF graph.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stanfordnlp/game-dialogue/issues/7#issuecomment-250364090, or mute the thread https://github.com/notifications/unsubscribe-auth/ABLcvwQGVJw-3zyDJEwe915hb-HMJ7h_ks5quzXCgaJpZM4KHP46 .

[hhexiy]

Yes.

[anushabala]

Thanks!!

On Wednesday, September 28, 2016, hhexiy notifications@github.com wrote:

Yes.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stanfordnlp/game-dialogue/issues/7#issuecomment-250364537, or mute the thread https://github.com/notifications/unsubscribe-auth/ABLcv2dkTYzHDlRvarCa0Y6NkKHAVYeMks5quzbngaJpZM4KHP46 .

[anushabala]

Hey! FYI, I just corrected a very minor typo in neural_session, you can just pull that whenever you get a chance :)

On Wed, Sep 28, 2016 at 8:56 PM, Anusha Balakrishnan < anusha.bala28@gmail.com> wrote:

Thanks!!

On Wednesday, September 28, 2016, hhexiy notifications@github.com wrote:

Yes.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stanfordnlp/game-dialogue/issues/7#issuecomment-250364537, or mute the thread https://github.com/notifications/unsubscribe-auth/ABLcv2dkTYzHDlRvarCa0Y6NkKHAVYeMks5quzbngaJpZM4KHP46 .

[hhexiy]

Cool. I suppose you'll run multiple controllers concurrently, right? I think this should work fine without extra work on the TF side, e.g. something like this: http://stackoverflow.com/questions/34419645/asynchronous-computation-in-tensorflow. But it's good to make sure the TF model will work as expected.. I'll do some simple tests and let you know.

[anushabala]

Yep, exactly. Oh that's a good point, I hope that it works well asymchronously! We can definitely do some tests once we have the website up and running, too.

On Thursday, September 29, 2016, hhexiy notifications@github.com wrote:

Cool. I suppose you'll run multiple controllers concurrently, right? I think this should work fine without extra work on the TF side, e.g. something like this: http://stackoverflow.com/questions/34419645/ asynchronous-computation-in-tensorflow. But it's good to make sure the TF model will work as expected.. I'll do some simple tests and let you know.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/stanfordnlp/game-dialogue/issues/7#issuecomment-250505489, or mute the thread https://github.com/notifications/unsubscribe-auth/ABLcvzCuYvgU9J73g_cX-LHprUfxpwwJks5qu90ggaJpZM4KHP46 .