philipperemy/keras-tcn

Multivariate Time Series Classification

mttbrt opened this issue · 4 comments

Hello, I found multi_length_sequences.py very useful and very effective to classify different length time series.

I'd like to ask if is it possible to classify multivariate time series (with different lengths) as well. I'm quite new in CNN and, specifically in TCN, so if you have a Python toy example or any hint I'd really appreciate it.

Thank you.

Hi @methK
Are you are talking about multi task learning ?

Thanks.

Hi @slateralus, thanks for your reply.
I think my problem is easier than multi task learning. I'd like to classify time series of different lengths as multi_length_sequences.py but these time series have a dimension greater than 1.

For example you can think of a multivariate time series as a series of length N where each point in time is an array of 3 elements (e.g. X, Y and Z values of a gyroscope).

Hi @methK

Input(batch_shape=(BATCH_SIZE, MAX_TIME_STEPS, EMBEDDING_LENGTH)
In your case EMBEDDING_LENGTH = 3

Thanks.

Thank you @slateralus, I did a quick test and it seems to work fine. I'll try more specific tests in the next hours and I will close the issue.

Thank you again.