dnc

dnc copied to clipboard

Dear author: We notice that in the paper, the memory is a 2D matrix (cell number, memory size]), but in pratice, the memory is a 3D tensor ([batch size, cell...

MrYaoH

can't run train.py

4

I ran `ipython3 train.py ` I got this:  I installed sonnet from this command `$ pip install dm-sonnet`

MustafaFayez

Upgrading repo to TF2

1

Upgrades repo from TF1 to TF2 while preserving the core design and logic. Quality of life: Formally separated unit testing Added requirements.txt Added precommit hooks for style Repeat Copy training...

solar464

I am very interested in this part of the experiments with Mini-SHRDLU, which inspired our current work. So, I would like to ask the author whether the code of this...

liyu199809

Hi, Are the _Family Tree_ and _London Underground_ datasets mentioned in the paper openly available? If so, where can I find them? Thanks, Thiviyan

thiviyanT

I wonder to know how can I create placeholders like dnc_core.initial_state(batch_size=n) with the batch_size n changed in trainning. > dnc_core = dnc.DNC(...) > initial_core_state = dnc_core.initial_state(batch_size=1) > core_state_placeholders = snt.nest.map(lambda...

hecong97

Is there any example implementation for block puzzle game using reinforcement learning?

koriavinash1

I would like to ask if there is a formal description of the multiple write head case, as this case is not presented in the original paper.

mammalwong

Convert to a package installable via pip

4

I just made this into a package that can be installed via pip. Thought it was a simple and useful enough change that you might be happy to merge it!...

edbordin