unit in movement_var

[samdeoxys1]

I'm a bit confused about the unit in movement_var from estimate_movement_var, as well as the resulting transition matrix using the RandomWalk transition type.

From the code it seems the unit of the returned movement_var is length / second. In make_state_transition, however, the transition probability is computed using the variance given by estinate_movement_var directly, without dividing it by the sampling frequency in the decoding problem. This means if my time bin for the decoded data is 0.1s, I'm still using a degree of smoothing as if my time bin were 1s.

Is this a correct interpretation of what the code is doing? Thanks!

[edeno]

Yeah, I think that's a bug in that code I should have fixed. I should have converted it to the correct time bin units. In practice in the paper I was going with a value that wasn't estimated, it was set to a reasonable value for replay (which is faster than the actual movement speed). I was just using this for the simulation. But your interpretation is correct.

[samdeoxys1]

Thanks! I found that sometimes after I converted it to the correct units, using the result from estimate_movement_var would lead to oversmoothing. I guess it would just need some extra tuning like you said.