Asking for multi signal support

sadransh commented 3 years ago

Thanks for sharing this nice library!

I wonder if it is possible to use your work for encoding multi signals.

I have 4 sensors where each gives 400 samples per instance ( time series, an equal number of samples.) So each of my objects has 4*400 samples ( and this have one label) . As an example, you can consider a dataset like the one used in this work: https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition

So in this, there is

INPUT_SIGNAL_TYPES = [
    "body_acc_x_",
    "body_acc_y_",
    "body_acc_z_",
    "body_gyro_x_",
    "body_gyro_y_",
    "body_gyro_z_",
    "total_acc_x_",
    "total_acc_y_",
    "total_acc_z_"
]

I would like to feed an autoencoder similar to the one done in the above-mentioned repo. So, the input tensor looks like this (7500, 128,9) => ( nb_of_objects, length of each sequence, nb_of_input_signals) in like this :

( this is object 0 of 7500 objects of train set)

[[ body_acc_x_[0], body_acc_y_[0], ... ,total_acc_y_[0],total_acc_z_[0]],
 [body_acc_x_[1], body_acc_y_[1], ... ,total_acc_y_[1],total_acc_z_[1] ] ,
 ...
 [body_acc_x_[127], body_acc_y_[127], ... ,total_acc_y_[127],total_acc_z_[127] ]]

Could you please tell me if it is possible to encode such signals using your work? If yes, please give me a little hint on how to do so.

In addition, I don't see any latent space visualization, Do you suggest any specific library to use for such a task?

shobrook commented 3 years ago

Hi Sadra, thanks for the question. sequitur supports what you're asking for, and you'll probably want to use the LSTM autoencoder (LSTM_AE). If you want to use quick_train, you'll first have to create a training set. Do this by creating a list of tensors, each with shape [sequence_length, num_signals]. Then plug in your training set and the model into quick_train, like so:

from sequitur.models import LSTM_AE
from sequitur import quick_train

train_set = ... # List of tensors, each with shape [sequence_length, num_signals]
encoding_dim = ... # Whatever you want the vector encoding size to be
encoder, decoder, _, _ = quick_train(LSTM_AE, train_set, encoding_dim=encoding_dim)

Please let me know if you have any further questions!

shobrook commented 3 years ago

Oh and check out https://projector.tensorflow.org/ if you want to visualize the latent space produced by the autoencoder.

shobrook commented 3 years ago

@sadransh Please let me know if this solves your issue so I can close it. Thanks!

sadransh commented 3 years ago

Thanks a lot. tbh in the meanwhile, I switched to Keras. However, the web-based projector you mentioned was really helpful to me. ( due to buggy behavior embedded projector for my case. )

As I review your readme I think it is clear that your library is able to do such. However, it was not clear to me as I was a beginner at the time I saw your work.

Thanks a lot for your help.

I might be able to create a notebook based on the HAR dataset using your library as a tutorial.

Are you willing to accept such a pull request?

shobrook commented 3 years ago

Glad I could help. And yes I'd be happy to accept that PR.

shobrook / sequitur

Asking for multi signal support #1