Resetting a counter

[jonahpearl]

Hi all — I'm trying to track a mouse in a simple arena (black mouse, white background), to detect when in it's a certain quadrant, and I want to add in some error robustness. The mouse has a tether, which causes only short-lived errors when the tether flops around, but as this is going to trigger optogenetic stimuli, I really want to avoid false positives. So, my current idea is to have a counter that counts frames when the mouse is in the quadrant, and set some threshold (say, 1 sec worth of frames) for allowing stimulation. Then if the mouse is ever detected out of the quadrant (truly or wrongly) the counter resets to 0 immediately.

I can't figure out how to implement this resetting behavior in Bonsai. My first thought was to use an Accumulator node and externalize its "value" property, but it turns out the Accumulator nodes don't have any properties at all. I think I might be able to just use a Python script module, but I'm going to run this at 120 fps and I don't want to have latency issues. Let me know if this is possible? Thank you!

place_pref_bonsai.zip

(Note: this is in Bonsai 2.8)

[banchan86]

Hi @jonahpearl there should be multiple ways of accomplishing it in bonsai without relying on external scripts, I am only showing one possible way here, not sure if its the best way or not. I simplified your workflow to something I could test with a computer mouse and also removed some unnecessary conversions.

I think the key is that you want to group your centroid readout into windows, and then evaluate each window using a SelectMany. You can use WindowTime or WindowCount for this purpose, and you can create overlapping or non overlapping windows depending on the latency you are looking for or if you are comfortable with just assessing each 1 second split independently (I don't think it takes up too much processing resources to evaluate sliding windows with a skip of just 1 element, but you should test that out)

Within the SelectMany you can filter out only the values that are True (when the mice is in the quadrant), and then count those values, if they are equal to the size of the window (your 1sec delay), then you can pass on a True to DigitalOutput. On the other hand, if there is a False value in the window, the count will not be equal and it will pass on a delay the moment the window with a False element pops up.

That being said, if you also know the behaviour of your tether artifact, it might be possible to skip over frames with that artifact. That would involve collecting some empirical data on possible constraints on area or kinematics (for instance, the largest binary region might be much bigger than the area for a regular size mouse, or the travelling velocity of the centroid might not be physically possible for a mouse). You could also try an online pose estimation package like Bonsai.Sleap. I believe the tracking is more accurate and resilient to tether artifacts. That might allow you to be less conservative about false positives and eliminate the minimum time condition.

Anyway let me know if the posted solution works for you, workflow attached! I kept the settings, so you can test it by moving your mouse over the top left corner.

#2343 place_pref_bonsai.zip

[jonahpearl]

Thank you, this is very helpful, I didn't know about the Window operators.

Two asides: 1) is there a good place to learn more about all the different operators in Bonsai / see more sophisticated examples? The docs feel very limited, and it's hard to learn to think in a "Bonsai-onic" (to borrow from pythonic") mindset.
2) I did notice a reference to the ROIActivity class in the state machine section of the docs, and using a simple binary threshold + the ROIActivityDetection with a high enough threshold turned out to work pretty well to exclude the tether artefacts (because the tether is thin and the mouse is quite large).

[banchan86]

Hey @jonahpearl glad that helped! Besides the tutorials (which you have already discovered), there is a 4-part video series on visual reactive programming in Bonsai which covers the reactive operators in depth. It can be found under the Learn section on the main bonsai website. Thanks for the feedback though, we are working on more accessible examples for each reactive operator that will go into the API page for each operator. Let me know if you also have other ideas about what resources we can further develop!