Hi,

I think the training progresses well, and there does'nt seem to be any improvement after 10,000 iterations. Sometimes I get nan s very early on, and have to restart, and I also got nans after 40,000 iterations.

May I ask how to serialise and save the DNC at checkpoints? I set a check point after 100 iterations, and got the following error,

Using CUDA.
Iteration 0/50000
	Avg. Logistic Loss: 0.6931
Iteration 50/50000
	Avg. Logistic Loss: 0.6674
Iteration 100/50000
	Avg. Logistic Loss: 0.4560

Saving Checkpoint ... Traceback (most recent call last):
  File "train.py", line 183, in <module>
    torch.save(ncomputer, f)
  File "/home/ajay/anaconda3/envs/rllab3/lib/python3.5/site-packages/torch/serialization.py", line 120, in save
    return _save(obj, f, pickle_module, pickle_protocol)
  File "/home/ajay/anaconda3/envs/rllab3/lib/python3.5/site-packages/torch/serialization.py", line 186, in _save
    pickler.dump(obj)
_pickle.PicklingError: Can't pickle <class 'memory.mem_tuple'>: attribute lookup mem_tuple on memory failed

PS - Sometimes I get nan s very early on, and have to restart, and I also got nans after 40,000 iterations. I've seen this very often with Neural Turing Machines, so I guess it's inherent with these type of things?

Using CPU.                   
Iteration 0/100000Traceback (most recent call last):
  File "train.py", line 177, in <module>
    output, _ = ncomputer(input_data)
  File "/home/jramapuram/.venv/lib/python2.7/site-packages/torch/nn/modules/module.py", line 206, in __call__
    result = self.forward(*input, **kwargs)
  File "../../neucom/dnc.py", line 108, in forward
    interface['erase_vector']
  File "../../neucom/memory.py", line 398, in write
    allocation_weight = self.get_allocation_weight(sorted_usage, free_list)
  File "../../neucom/memory.py", line 144, in get_allocation_weight
    flat_unordered_allocation_weight.cpu()
  File "/home/jramapuram/.venv/lib/python2.7/site-packages/torch/autograd/variable.py", line 629, in scatter_
    return Scatter(dim, True)(self, index, source)
RuntimeError: a leaf Variable that requires grad has been used in an in-place operation.

Hi thanks for sharing this, it's really cool that you managed to get it working :+1:

I got an error though when trying to run the repo,

Iteration 0/100000Traceback (most recent call last):
  File "train.py", line 155, in <module>
    loss.backward()
  File "/home/ajay/anaconda3/envs/pyphi/lib/python3.6/site-packages/torch/autograd/variable.py", line 158, in backward
    self._execution_engine.run_backward((self,), (gradient,), retain_variables)
  File "/home/ajay/anaconda3/envs/pyphi/lib/python3.6/site-packages/torch/autograd/variable.py", line 201, in _do_backward
    "leaf variable was used in an inplace operation"
AssertionError: leaf variable was used in an inplace operation

I've seen this before but I can't remember how to fix it? Can you suggest anything please?

Also maybe there's a typo on the "how to run" part of the cover page for the repo, I think it should be ,

python tasks/Copy/train.py

Thanks a lot for your help :+1:

mldl@mldlUB1604:~/ub16_prj/pytorch-NeuCom/tasks/Copy$ python train.py Using CUDA. Iteration 0/100000Traceback (most recent call last): File "train.py", line 177, in output, _ = ncomputer(input_data) File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 206, in call result = self.forward(*input, **kwargs) File "../../neucom/dnc.py", line 108, in forward interface['erase_vector'] File "../../neucom/memory.py", line 384, in write lookup_weight = self.get_content_address(memory_matrix, key, strength) File "../../neucom/memory.py", line 78, in get_content_address cos_dist = cosine_distance(memory_matrix, query_keys) File "../../neucom/utils.py", line 154, in cosine_distance memory_norm = torch.norm(memory_matrix, 2, 2, keepdim=True) TypeError: norm() got an unexpected keyword argument 'keepdim' mldl@mldlUB1604:~/ub16_prj/pytorch-NeuCom/tasks/Copy$

(VENVpytorch2) mldl@mldlUB1604:~/ub16_prj/pytorch-NeuCom/tasks/Copy$ python train.py Using CUDA. Iteration 0/100000Traceback (most recent call last): File "train.py", line 177, in output, _ = ncomputer(input_data) File "/home/mldl/ub16_prj/VENV_host/VENV_pytorch2/VENVpytorch2/local/lib/python2.7/site-packages/torch/nn/modules/module.py", line 206, in call result = self.forward(*input, **kwargs) File "../../neucom/dnc.py", line 108, in forward interface['erase_vector'] File "../../neucom/memory.py", line 384, in write lookup_weight = self.get_content_address(memory_matrix, key, strength) File "../../neucom/memory.py", line 78, in get_content_address cos_dist = cosine_distance(memory_matrix, query_keys) File "../../neucom/utils.py", line 154, in cosine_distance memory_norm = torch.norm(memory_matrix, 2, dim=True) File "/home/mldl/ub16_prj/VENV_host/VENV_pytorch2/VENVpytorch2/local/lib/python2.7/site-packages/torch/autograd/variable.py", line 596, in norm return Norm(p, dim)(self) File "/home/mldl/ub16_prj/VENV_host/VENV_pytorch2/VENVpytorch2/local/lib/python2.7/site-packages/torch/autograd/_functions/reduce.py", line 192, in forward output = input.norm(self.norm_type, self.dim) TypeError: norm received an invalid combination of arguments - got (int, bool), but expected one of:

• no arguments
• (float p)
• (float p, int dim) didn't match because some of the arguments have invalid types: (int, bool)

(VENVpytorch2) mldl@mldlUB1604:~/ub16_prj/pytorch-NeuCom/tasks/Copy$

Hey, In function get_allocation_weight() in memory.py, cumprod is used, but as far as I know, this operation does not have autograd support yet and is still in PR: https://github.com/pytorch/pytorch/pull/1439, so I'm really confused as how is the backward passed through this operation for you? Is the resulting variables not in the computation graph? Thanks in advance! Really nice repository:)

Hi @ypxie,

I just saw DeepMind released their Tensorflow/Sonnet implementation of the DNC,

https://github.com/deepmind/dnc

I was wondering if you're interested in re-looking at this project to see if we can compare it with DMs code? I spent quite a lot of time previously on your implementation, be a lot of fun to start working on it again :+1:

Kind regards,

Ajay