saif-mahmud / self-attention-har Goto Github PK

maybe it should be data.append([float(x) / 1000 for x in elem[1:]])?

Hi, when i follow your instruction, i got OSError: Unable to open file (unable to open file: name = 'data/processed/opportunity.h5', errno = 2, error message = 'No such file or directory', flags = 0, o_flags = 0), how can I fix it ? thanks

You used the train_test_split from sklearn for Skoda dataset. But you did not set the shuffle parameter in train_test_split function. It will break all temporal dependencies within the data. The data is reordered.

Then you applied the sliding window function on the reordered data.

Epoch 00013: early stopping

why this happens?

I ran for 75 epochs

@thTonmoy @saif-mahmud could you please re-consider taking a quick look at #3? It will probably not take much time.

Basically, if we fix the pre-processing step as clarified in the above-referenced issue, then the model's performance turns into a disaster.

I was only able to replicate your paper's result with the current error in the pre-processing step. This error consists in that it includes the label (ground-truth) in the training/validation/testing sets. But this doesn't make any sense.

How can I get the sensor attention weight for each category and visualize it