Modification on Dataset

To handle real data we need some modifications on the current implementation to train with data from multiple rosbags.

We discussed two options:

1. Make the real data fit to the shape of synth data, means to successively load the bags and create a dataset similiar to the synth data shape (num_bags, num_samples, num_features)
2. To read the bags and concatenate all samples within one entry, resulting in a shape similiar to (num_samples*num_bags, num_features) like it can be seen here 19e96697. Following it's needed to modify dataset_iter_multi.py and the v_map fn.

@Update: After discussing this thread with Kevin, we:

• Modify the slide v_map fn
• Convert to np.ndarray after sliding due to uniform size of the data
• Write better concatenate part without addressing each attribute