Modification of the experience during training

[glezmartin]

Hello everyone,

I am trying to train a model with avalanche with a dataset created through avalanche tools. During training I need to alter the experience delivered to the model by adding a m-dimensional tensor to each of the n-dimensional observations of the batch of that experience. I have tried to access the information contained in the experience and I have not been able to do so. How can I make this modification on the experience?

The main idea would be to do this operation

$h_{t+1}, s_{t+1} = LSTM(o^i_t + f_t , h_{t}, s_{t})$

where $o^i_t$ would be the i-th sample in the experience and $f_t$ would be the computed tensor that would be concatenated to $o^i_t$ resulting in a experience with n+m dimension samples.

Thank you!