I am trying to train your model to learn tactile representations with data from my sensor. I see that in you replay_buffer.zarr data, you are also saving episode ends. Is it necessary if I'm only interesed in the tactile representation task? If I just save one number that matches the shape of my dataset, can I train the model without problems?
Yes. One episode is basically a chunk of data with shape (T, data_shape). Here T could be the total number of your images and data_shape is the size of your images.
Hi,
I am trying to train your model to learn tactile representations with data from my sensor. I see that in you replay_buffer.zarr data, you are also saving episode ends. Is it necessary if I'm only interesed in the tactile representation task? If I just save one number that matches the shape of my dataset, can I train the model without problems?
Thanks in advance.