Error when 'learning_starts' Parameter is Smaller Than 'topk' in Notebook

[KoOBaALT]

I encountered an error in a notebook when the learning_starts parameter is smaller than the topk value. Specifically, the error occurs in the following code snippet:

lge.explore(num_timesteps)

It seems related to the default learning_starts parameter of lge being set to 100:

def __init__(... , learning_starts: int = 100, ...)

This causes an error because when the element in topk is smaller than the parameter k, it raises an error:

   cdist = torch.cdist(x, samples)
   dist_to_kst = cdist.topk(k, largest=False)[0][:, -1]

I was thinking of replacing k with:

k = min(1000, x.shape[0]

However, I'm uncertain if there's a specific intention behind hardcoding k = 1000. Could this be a viable solution, or is there a particular reason for the current implementation?