Open ZZysh opened 2 years ago
sronilsson
commented
2 years ago Hi @ZZysh! You're right, hyperparameters can be tricky, but I can give some pointers. To avoid giving you some generic answers, can you tell me what types of animals, how many animals, and which body-parts you are tracking?
ZZysh
commented
2 years ago thank you so much for so soon reply! I have 2 experimental conditions(or session) (1) first, there are 2 c57 mice（black and one of them the back fur are shaved and can be manual discrimination） in a square box and recorded by a infrared camera i use deeplabcut to analyze the pose estimation and there are 8 body points used as suggested. however the tail rear is difficult to identify and I prefer to use the 7 points you suggested. (2)the other condition is only one c57 mice in the box
I wanna to cluster the behavior, especially allogroming, rearing, moving, and static(I mean motionless)
thanks again for rapid reply!
---- 回复的原邮件 ---- | 发件人 | Simon @.> | | 日期 | 2022年04月21日 23:39 | | 收件人 | @.> | | 抄送至 | @.**@.> | | 主题 | Re: [sgoldenlab/simba] Seeking help for some parameter setting (Issue #177) |
Hi @ZZysh! You're right, hyperparameters can be tricky, but I can give some pointers. To avoid giving you some generic answers, can you tell me what types of animals, how many animals, and which body-parts you are tracking?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
sronilsson
commented
2 years ago Cheers!
For outlier correction - I'd start with what was used in this paper https://www.biorxiv.org/content/10.1101/2020.04.19.049452v2.full.pdf. ...movement criterion to 0.7, and location criterion to 1.5. If you know your tracking is good and contains few outliers, there is no need to perform outlier correction, and you can slick the Skip button. An alternative approach, is to smooth the data, documented HERE and example video HERE which should also help with outliers.
For the hyperparameters, you can start with the ones in THIS FILE. You can download this file, and then load this into the SimBA GUI as documented HERE so you don't have to fill in any hyperparameters to start. If they don't work as well as anticipated, I recommend beginning by titrating the undersample ratio.
ZZysh
commented
2 years ago thanks a lot I will try
---- 回复的原邮件 ---- | 发件人 | Simon @.> | | 日期 | 2022年04月22日 02:55 | | 收件人 | @.> | | 抄送至 | @.**@.> | | 主题 | Re: [sgoldenlab/simba] Seeking help for some parameter setting (Issue #177) |
Cheers!
For outlier correction - I'd start with what was used in this paper https://www.biorxiv.org/content/10.1101/2020.04.19.049452v2.full.pdf. ...movement criterion to 0.7, and location criterion to 1.5. If you know your tracking is good and contains few outliers, there is no need to perform outlier correction, and you can slick the Skip button. An alternative approach, is to smooth the data, documented HERE and example video HERE which should also help with outliers.
For the hyperparameters, you can start with the ones in THIS FILE. You can download this file, and then load this into the SimBA GUI as documented HERE so you don't have to fill in any hyperparameters to start. If they don't work as well as anticipated, I recommend beginning by titrating the undersample ratio.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
ZZysh
commented
2 years ago Hi Simon, i wanna to use the gpu to accelerate the video trim function(Ffmepg) assemble in Simba, because I have a long video. could you give me some advise? thanks a lot and looking forward your reply
thanks a lot I will try
---- 回复的原邮件 ---- | 发件人 | Simon @.> | | 日期 | 2022年04月22日 02:55 | | 收件人 | @.> | | 抄送至 | @.**@.> | | 主题 | Re: [sgoldenlab/simba] Seeking help for some parameter setting (Issue #177) |
Cheers!
For outlier correction - I'd start with what was used in this paper https://www.biorxiv.org/content/10.1101/2020.04.19.049452v2.full.pdf. ...movement criterion to 0.7, and location criterion to 1.5. If you know your tracking is good and contains few outliers, there is no need to perform outlier correction, and you can slick the Skip button. An alternative approach, is to smooth the data, documented HERE and example video HERE which should also help with outliers.
For the hyperparameters, you can start with the ones in THIS FILE. You can download this file, and then load this into the SimBA GUI as documented HERE so you don't have to fill in any hyperparameters to start. If they don't work as well as anticipated, I recommend beginning by titrating the undersample ratio.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
sronilsson
commented
2 years ago Hi @ZZysh - it is not something that is supported in SimBA and you'd have to do code up yourself outside the SimBA environment. It does seem possible but you'd have to tinker with the ffmpeg installation... https://www.cyberciti.biz/faq/how-to-install-ffmpeg-with-nvidia-gpu-acceleration-on-linux/
If you do give it a try, please let me know how it goes and if you manage to get it working.
ZZysh
commented
2 years ago thanks I will try it latter and I will tell you after i do it thanks a lot
---- 回复的原邮件 ---- | 发件人 | Simon @.> | | 日期 | 2022年04月22日 20:08 | | 收件人 | @.> | | 抄送至 | @.**@.> | | 主题 | Re: [sgoldenlab/simba] Seeking help for some parameter setting (Issue #177) |
Hi @ZZysh - it is not something that is supported in SimBA and you'd have to do code up yourself outside the SimBA environment. It does seem possible but you'd have to tinker with the ffmpeg installation... https://www.cyberciti.biz/faq/how-to-install-ffmpeg-with-nvidia-gpu-acceleration-on-linux/
If you do give it a try, please let me know how it goes and if you manage to get it working.
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
Hi, there are some question, i wanna to seeking your help (1) Fistly, what is the recommend outlier correct criterion between movement and location, is it 1 for (movement) and 1.5 for (location) ok? (2) Secondly, what is the recommended hyperparameters used in the RF(ramdom forest modle). to be honest , it is unfamiliar for me and it seems difficut. thankyou for you attention!