Why log var network is defined as a single parameter(its expression ability seems being limited compared with a neural network) and initialized with log(0.1) ? https://github.com/wyndwarrior/Sectar/blob/2e0d67481a/exps/swimmer/swimmer_explore.py#L92

I notice that in gaussian_network.py, you use min_log_var in GaussianNetwork class to ensure the variance of z distribution not too low. However, not all subclass use this variable(e.g. GaussianBidirectionalNetwork). I am wondering why we need this variable and how this variable effect training?

Thank you :)

I don't understand why you use x[:, self.step_dim:] rather than x[:, :self.step_dim] to compute MSE. As far as I know, x[:, :self.step_dim] is the observation for the total trajectory. What's the meaning of x[:, self.step_dim:]?
https://github.com/wyndwarrior/Sectar/blob/master/traj2vec/algos/vae_bc.py#L90

Thanks!

Hi,
I'm confused with this mpc code. In both outer loop and inner loop, 'i' is used as index. Does this affect the training?

@wyndwarrior

I just want to get this code running but I am lost with the scripts. Which script is the entry one to train?

Hi,
Amazing work! I was working towards generating vectors from GPS trajectories, I have a dataset similar to GeoLife, i.e. it contains userID, timestamp, latitude and longitude in a dataframe. Can you guide on how to use this code to generate vectors of the respective trajectories?