Maghoumi/pytorch-softdtw-cuda

Issues

• It is support time-serise Classification task?

  #34 opened 5 months ago by CAI23sbP
  2

• why i get a negative loss value?

  #32 opened 10 months ago by FZH1802
  4

• Grid size (4) < 2 * SM count (164) will likely result in GPU under utilization due to low occupancy.

  #21 opened 2 years ago by 18445864529
  1

• output the alignments of two sequences

  #31 opened a year ago by XIDIANPQZ
  2

• Incorrect Batch size for euclidian dist using normalization

  #30 opened a year ago by denisbeslic
  0

• NvvmSupportError: libNVVM cannot be found.

  #28 opened 2 years ago by zxg-code
  0

• sympy error

  #29 opened a year ago by gg4u
  0

• Negative value

  #27 opened 2 years ago by NingkangYang
  1

• CUDA_ERROR_INVALID_VALUE

  #26 opened 2 years ago by Cram3r95
  3

• Gradients are gone when moving the code to CUDA

  #25 opened 2 years ago by mariaalfaroc
  0

• package in pypi or conda

  #20 opened 2 years ago by toinsson
  2

• Regarding GPU memory footprint

  #19 opened 2 years ago by netw0rkf10w
  1

• The loss value of softDTW is inf when choosing bandwidth values

  #18 opened 3 years ago by educationunion
  1

• GPU mem still be affected when using CPU Soft DTW

  #13 opened 3 years ago by v-nhandt21
  2

• Compare with Cython implement

  #12 opened 3 years ago by v-nhandt21
  2

• Example fails on GPU

  #14 opened 3 years ago by weidenka
  7

• [Feature Request] Compatibility with Pytorch packed sequences

  #16 opened 3 years ago by parthatom
  1

• can DTW be negative?

  #10 opened 3 years ago by yongjunshin
  1

• AssertionError when dims increase

  #15 opened 3 years ago by tricky61
  0

• Batch of variable sequence length

  #11 opened 3 years ago by daidedou
  2

• Does this implementation support distributed training in Pytorch?

  #7 opened 4 years ago by taolusi
  4

• loss drop below 0?

  #9 opened 3 years ago by dyt0414
  1

• soft-DTW loss became inf when setting bandwidth bigger than 0

  #8 opened 3 years ago by LuckerYi
  2

• Does .backward() contain grads for both sequences?

  #5 opened 4 years ago by janvainer
  2

• [Feature request] Add backtracking/compatibility with other tools

  #6 opened 4 years ago by janvainer
  1

• Value for bandwitdh pruning?

  #4 opened 4 years ago by hadaev8
  3

• Distance function and comparison with no warping

  #3 opened 4 years ago by cfrancesco
  3

• Steps chosen / choice penalty

  #2 opened 4 years ago by cfrancesco
  3

• *obj* doesn't implement the cuda array interface. at cuda.as_cuda_array(D.detach())

  #1 opened 4 years ago by KimUyen
  2