emperor
emperor copied to clipboard
Adding in kernel density estimation
One thing that I think could potentially really clarify the the PCoA plots is to have some kernel density estimators. This could not only allow the user to visualize where the most points are clustered, but would definitely ease visualizing classification in the future.
We can start off using a gaussian kernel density estimator. From my experience, kernel density estimation is pretty fast, so we don't need external preprocessing.
Are you thinking this could be part of the graphical user interface or a separate script?
On (Mar-07-15|16:31), mortonjt wrote:
One thing that I think could potentially really clarify the the PCoA plots is to have some kernel density estimators. This could not only allow the user to visualize where the most points are clustered, but would definitely ease visualizing classification in the future.
We can start off using a gaussian kernel density estimator. From my experience, kernel density estimation is pretty fast, so we don't need external preprocessing.
Reply to this email directly or view it on GitHub: https://github.com/biocore/emperor/issues/360
I think this should be a flag in make_emperor.py.
I'm not sure how that would be represented on the plot, are we thinking on adding a 3D surface or what exactly?
On (Mar-09-15| 9:33), Antonio Gonzalez wrote:
I think this should be a flag in make_emperor.py.
Reply to this email directly or view it on GitHub: https://github.com/biocore/emperor/issues/360#issuecomment-77889649
I was actually thinking that it should be part of the graphical interface - the user can display and clear the kernel densities on the fly.
I originally thinking that these kernel densities could be displayed similar the jack-knifed plots that we have - but have a contour of gradients to visualize the density.
Here is an example of something similar to what I have in mind.
This will be difficult - may take a bit of tinkering before a PR is issued.
That will be really cool but note that this will require to have the client (web interface) issue python jobs.
That would be really awesome! Certainly worth adding.
On Mar 9, 2015, at 12:45 PM, mortonjt <[email protected]mailto:[email protected]> wrote:
I was actually thinking that it should be part of the graphical interface - the user can display and clear the kernel densities on the fly.
I originally thinking that these kernel densities could be displayed similar the jack-knifed plots that we have - but have a contour of gradients to visualize the density.
Here is an examplehttp://www.star.bris.ac.uk/%7Embt/topcat/v4_graphics.html of something similar to what I have.
This will be difficult - may take a bit of tinkering before a PR is issued.
— Reply to this email directly or view it on GitHubhttps://github.com/biocore/emperor/issues/360#issuecomment-77927401.
Well, another thing that would be really cool that requires the same ability would be the highly desirable task of re-trigerring pcoa with just the selected samples so it's worth exploring whether it's feasible.
On Mar 9, 2015, at 12:47 PM, Antonio Gonzalez <[email protected]mailto:[email protected]> wrote:
That will be really cool but note that this will require to have the client (web interface) issue python jobs.
— Reply to this email directly or view it on GitHubhttps://github.com/biocore/emperor/issues/360#issuecomment-77927730.