mcren88 / differentiable_neural_computer Goto Github PK

• My implementation of Differentiable Neural Computer (DNC) in PyTorch.
• DNC is introduced in the paper Hybrid computing using a neural network with dynamic external memory.
• Currently, bAbI Question Answering task and Pattern Copy task is implemented for CPU and GPU both.
• Although I have tested the code thoroughly, bugs may persist. In that case you are encouraged to report them.

The code is written in Python 3.6 using PyTorch 1.1.0 in Ubuntu 18.04 Operating System.

NumPy
PyTorch

python3 train.py opt1 opt2

Options: 
opt1: 1). 1 for Copy_Task
      2). 2 for bAbI_Task

opt2: 1). GPU to run code on GPU
      2). CPU to run code normally

python3 test.py opt1 opt2 opt3 opt4

Options: 
opt1: 1). 1 for Copy_Task
      2). 2 for bAbI_Task

opt2: 1). GPU to run code on GPU
      2). CPU to run code normally

opt3: Last Epoch number till the model was trained (Not Applicable for Copy Task. Any value is fine)

Opt4: Last Batch Number till the model was trained 

1). https://github.com/loudinthecloud/pytorch-ntm
2). https://github.com/bgavran/DNC