nupic.torch package

• nupic.torch.functions package
  • nupic.torch.functions.k_winners
    • kwinners()
    • kwinners2d()
• nupic.torch.models package
  • nupic.torch.models.sparse_cnn
    • GSCSparseCNN
    • GSCSuperSparseCNN
    • MNISTSparseCNN
    • gsc_sparse_cnn()
    • gsc_super_sparse_cnn()
• nupic.torch.modules package
  • nupic.torch.modules.flatten
    • Flatten
      • Flatten.forward()
  • nupic.torch.modules.k_winners
    • KWinners
      • KWinners.extra_repr()
      • KWinners.forward()
      • KWinners.update_duty_cycle()
    • KWinners2d
      • KWinners2d.entropy()
      • KWinners2d.extra_repr()
      • KWinners2d.forward()
      • KWinners2d.update_duty_cycle()
    • KWinnersBase
      • KWinnersBase.entropy()
      • KWinnersBase.extra_repr()
      • KWinnersBase.max_entropy()
      • KWinnersBase.update_boost_strength()
      • KWinnersBase.update_duty_cycle()
    • update_boost_strength()
  • nupic.torch.modules.prunable_sparse_weights
    • PrunableSparseWeightBase
      • PrunableSparseWeightBase.off_mask
    • PrunableSparseWeights
    • PrunableSparseWeights2d
  • nupic.torch.modules.sparse_weights
    • HasRezeroWeights
      • HasRezeroWeights.rezero_weights()
    • SparseWeights
      • SparseWeights.rezero_weights()
    • SparseWeights2d
      • SparseWeights2d.rezero_weights()
    • SparseWeightsBase
      • SparseWeightsBase.extra_repr()
      • SparseWeightsBase.forward()
      • SparseWeightsBase.bias
      • SparseWeightsBase.weight
      • SparseWeightsBase.weight_sparsity
    • normalize_sparse_weights()
    • rezero_weights()

nupic.torch.compatibility

upgrade_to_masked_sparseweights(state_dict)

Returns a new state dict with any “zero_weights” tensors converted to “zero_mask” tensors. (The “zero_weights” was a list of indices of zeroes in the weight tensor.)

nupic.torch.duty_cycle_metrics

binary_entropy(x)

Calculate entropy for a list of binary random variables.

Parameters

x – (torch tensor) the probability of the variable to be 1.

Returns

entropy: (torch tensor) entropy, sum(entropy)

max_entropy(n, k)

The maximum entropy we could get with n units and k winners.