mqwfrog / mhccl Goto Github PK

Good Jobs！
Hello, author.After reading your code, I have a question that the model designed specifically for multivariate time series? Does that mean can't run on a univariate time series.

Hi, I'm trying to replicate your algorithm, but I've found a value error when I try to run the wisdm dataset:

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.

This error is raised at:

augmentations.py, line 41, in permutation
    warp = np.concatenate(np.random.permutation(splits)).ravel()
  File "mtrand.pyx", line 4703, in numpy.random.mtrand.RandomState.permutation

Do you know how I can solve it?

Hi,

I am attempting to set up the 7 datasets as listed in your paper, and I've encountered several issues. I was able to work with some datasets from UCI and UEA, but others presented challenges.

Successfully Accessed Datasets

PenDigits, EigenWorms, FingerMovements

• For these UEA datasets, there were no issues.

HAR

• The HAR dataset was downloaded from the UCI repository.
• After preprocessing, the HAR dataset matches the number of features and time length as expected.
  

Issues with Specific Datasets

Epilepsy

• There seems to be a confusion whether the Epilepsy dataset belongs to UCI or UEA, as discussed in this issue.
• When using the Epilepsy dataset from UEA, there is a mismatch in the number of features and the time length expected as per the paper. An illustration of the issue is shown below:
  

WISDM

• I encountered issues when trying to preprocess the dataset (downloaded from the UCI repository) with the provided code. The code references a specific file named WISDM_ar_v1.1_raw.txt, which was not available in the downloaded files.
• In an attempt to locate the correct version of the dataset, I searched Google and found a potential match on Kaggle.
• Despite finding this file and attempting to preprocess it, there remains a mismatch in the number of features and the time length compared to what is expected based on the information in the paper.
  

SHAR

• I am unable to locate the SHAR dataset. While I did find a related dataset on Kaggle, it does not contain a .mat file, which seems to be required for the preprocessing code provided.

I would appreciate any assistance or guidance on accessing and preprocessing these datasets correctly.

Thank you.

Hello, I cannot reproduce good results on wisdm with the following parameter settings, could you share the exact parameter setting for wisdm?

python main.py \
--dataset_name wisdm \
--lr 0.03 \
--batch_size 128 \
--mlp --cos \
--layers 3 \
--posi 2 \
--negi 100 \
--posp 3 \
--negp 4 \
--gpu 7 --epochs 150
--mask_layer0 \
--mask_mode 'mask_threshold'

Within your implementation, you guys used conv2d in ResNet implementation. Just out of curiosity, why don't you use conv1d, which is the norm in the time-series domain? Thanks.

Within your implementation, you guys used conv2d in ResNet. Just out of curiosity, why don't you use conv1d, which is the norm in the time-series domain? Thanks.