siddharth-agrawal / neural-tensor-network Goto Github PK

I am getting below error when trying to run python neuralTensorNetwork.py

File "neuralTensorNetwork.py", line 753, in <module>
    neuralTensorNetwork()
  File "neuralTensorNetwork.py", line 733, in neuralTensorNetwork
    args = (data, 1,), method = 'L-BFGS-B', jac = True, options = {'maxiter': batch_iterations})
  File "/Users/khurram/anaconda3/envs/tf2/lib/python2.7/site-packages/scipy/optimize/_minimize.py", line 487, in minimize
    callback=callback, **options)
  File "/Users/khurram/anaconda3/envs/tf2/lib/python2.7/site-packages/scipy/optimize/lbfgsb.py", line 335, in _minimize_lbfgsb
    f, g = func_and_grad(x)
  File "/Users/khurram/anaconda3/envs/tf2/lib/python2.7/site-packages/scipy/optimize/lbfgsb.py", line 285, in func_and_grad
    f = fun(x, *args)
  File "/Users/khurram/anaconda3/envs/tf2/lib/python2.7/site-packages/scipy/optimize/optimize.py", line 292, in function_wrapper
    return function(*(wrapper_args + args))
  File "/Users/khurram/anaconda3/envs/tf2/lib/python2.7/site-packages/scipy/optimize/optimize.py", line 63, in __call__
    fg = self.fun(x, *args)
  File "neuralTensorNetwork.py", line 219, in neuralTensorNetworkCost
    entity_vectors_e1 = entity_vectors[:, e1.tolist()]
IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices

You have this function, StackToParams where you seem to take all the models' parameters and roll them into one parameter, which seems moderately expensive. The only place it seems like you us this representation of your parameters is that allows you to say this, which is clean, I admit. But this seems like a lot of overhead to keep the code a little cleaner. And for this cleanliness, I feel like you've lost some readability. Am I missing something? Is there some other reason you chose to wrap all those parameters together?

Hi Guys!
I don't understand why the following line of NeuralTensorNetwork.py works well.

wrong_filter = (score_pos + 1 > score_neg)[0]

I think you should subtract 1 from ``score_pos''.

wrong_filter = (score_pos - 1 > score_neg)[0]

What is 'wordIndices.p' file and how is it created ?

File "neuralTensorNetwork.py", line 219, in neuralTensorNetworkCost
entity_vectors_e1 = entity_vectors[:, e1.tolist()]
IndexError: only integers, slices (:), ellipsis (...), numpy.newaxis (None) and integer or boolean arrays are valid indices

,

Hello,

I think the acitviation function for calculating the score of a tripple after the training is missing.
https://github.com/siddharth-agrawal/Neural-Tensor-Network/blob/master/neuralTensorNetwork.py#L428-L434 and also in getPredictions().

dev_scores[i, 0] += U[rel][k, 0] * (np.dot(entity_vector_e1.T, np.dot(W[rel][:, :, k], entity_vector_e2)) +
np.dot(V[rel][:, k].T, entity_stack) + b[rel][0, k])

I think it should be like this:
test_score += U[rel][k, 0] * self.activationFunction((np.dot(entity_vector_e1.T, np.dot(W[rel][:, :, k], entity_vector_e2)) +
np.dot(V[rel][:, k].T, entity_stack) + b[rel][0, k]))

What do you think?

Hi,
When I run the neuralTensorNetwork.py file, I have encountered this problem:
I use Python 3.6.

Traceback (most recent call last):
File "neuralTensorNetwork.py", line 764, in
neuralTensorNetwork()
File "neuralTensorNetwork.py", line 710, in neuralTensorNetwork
word_indices, num_words = getWordIndices('wordIndices.p')
File "neuralTensorNetwork.py", line 591, in getWordIndices
word_dictionary = pickle.load(open(file_name, 'rb'))
UnicodeDecodeError: 'ascii' codec can't decode byte 0xdf in position 0: ordinal not in range(128)

Does anyone know how to solve it?

Thanks.