Kilosort support and performance

[Gregory Telian]

Hi fellow klustaviewas!

I've been using the KlustaSuite and I absolutely love it. It has allowed me to easily and
quickly sort the large amounts of data I collect with silicone electrodes.

I've heard that there might be a transition to Kilosort and that it is much faster and more
accurate that Klustakwik. Has anyone tried it out?

I'm hesitant to use it, even if it is better, because it doesn't seem as well supported as the KlustaSuite
(i.e. No google group, not many people using it). It also hasn't been published yet like KK.

Can anyone share what they know about Kilosort? Specifically, does it take electrode geometry into account like KK? How difficult is it to setup? Why does it not use the kwik file format? Will
KK be replaced by it (i.e. KK will not be under active development)?

tl;dr
Is it worth switching to Kilosort or should I keep using the KlustaSuite? What are people's
thoughts and experiences with Kilosort?

Thank you!!!
I look forward to people's replies!

Greg Telian

[Harris, Kenneth]

Hi Greg,

There will indeed be a transition to Kilosort and phy in the long term. We are gradually transitioning to kilosort within our own lab, and the users who have adopted it so far have been very happy with it.

You are correct that kilosort is not yet publically supported. It is for now provided on an "as-is" basis: although some documentation is provided, we cannot help you install it on your system, configure the parameter settings, or make sense of the output. We will eventually transition to supporting kilosort to the same level as KlustaSuite, but first it has to go through extensive beta testing in a smaller community.

So my advice would be: if you are happy with KlustaSuite, stick to this for now. The biggest advantage of kilosort is that it runs much faster on very large electrode arrays. Unless you are finding execution speed to be a serious problem, there is no need to transition until kilosort becomes more stable.

Kilosort does not explicitly take into account electrode geometry, but because of other changes in the algorithm it can learn this implicitly from the data. You can read more about the algorithm at http://biorxiv.org/content/early/2016/06/30/061481.

Regarding file formats: we are moving on from the kwik format because of the general opinion of users and developers that HDF5 did not provide any advantage over a directory of individual files. The new format used by kilosort is designed to work with a forthcoming proposal for worldwide neurophysiology data indexing and sharing. More details on this to follow.

All the best,
Kenneth.

--
To subscribe, send email to klustaviewa...@googlegroups.com
---
You received this message because you are subscribed to the Google Groups "KlustaViewas" group.
To unsubscribe from this group and stop receiving emails from it, send an email to klustaviewas...@googlegroups.com.
To post to this group, send email to klusta...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Gregory I. Telian]

Hi Kenneth,

Thank you so much for such a detailed response!!!

All of that sounds incredibly exciting. I'm really looking forward to the future transition both with Kilosort and with the new neurophys data format!

Thank you and the entire Klusta team for producing such useful and amazing software! It is so so much better than the sorters I've used in the past. If I could be of any help during the beta testing please let me know.

Cheers,

Greg 

[Nick Steinmetz]

Hi Greg (and everyone), 

If you do wish to beta test the newer Kilosort/phy, with the understanding that it is still under development as Kenneth says, you can find Kilosort here and phy here, as well as what should be a user-friendly documentation of the new gui and file format here. Problems with Kilosort can be posted as issues on Kilosort's github page, and questions about phy can be addressed to the phy-users google group, or its issues page on github. We would love to hear any feedback you have. 

Happy sorting,

Nick

Greg 

To subscribe, send email to klustaviewas+subscribe@googlegroups.com

---
You received this message because you are subscribed to the Google Groups "KlustaViewas" group.

To unsubscribe from this group and stop receiving emails from it, send an email to klustaviewas+unsubscribe@googlegroups.com.

To post to this group, send email to klusta...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
To subscribe, send email to klustaviewas+subscribe@googlegroups.com

---
You received this message because you are subscribed to the Google Groups "KlustaViewas" group.

To unsubscribe from this group and stop receiving emails from it, send an email to klustaviewas+unsubscribe@googlegroups.com.