contact_map
contact_map copied to clipboard
dwhswenson
Enhancement: ContactTrajectory class (with movies!)
It might be nice to add a ContactTrajectory class. This would store a separate (single frame) contact map for each time slice. Then you could do interesting things like create a ContactFrequency based on a set of frames determined after the fact. You could also create movies of how the contact map evolves in time (these should be based on some windowing into ContactFrequency instances; the window can be 1 if the user wants).
Relevant matplotlib docs on animation:
- https://matplotlib.org/api/animation_api.html
- https://matplotlib.org/gallery/animation/dynamic_image.html
Animations that could be generated in real time would be impressive, but I'd be just as happy with something that pre-generates the images and makes a movie out of them, like ArtistAnimation does.
Note that, to use a rolling window to generate the contact frequency plots for each movie frame, we already have add_contact_frequency and subtract_contact_frequency methods. These affect the absolute histogram count, unlike ContactDifference, which compares the after-normalization (to number of frames) contact frequencies. We'd still output a separate plot at each time step, but the underlying data would be cheaper to calculate.
+1 for the idea.
We'd still output a separate plot at each time step, but the underlying data would be cheaper to calculate.
Which one do you expect to be cheaper, using the methods or the ContactDifference?
Which one do you expect to be cheaper, using the methods or the
ContactDifference?
Not sure what you mean here... these calculate different things.
If you have freq_1 where a given contact is present 20/25 frames (80%), and freq_2 where that contact is present 1/2 frames (50%), then ContactDifference(freq_1, freq_2) gives a result of 30% for that contact (80% - 50%). On the other hand, freq_1.subtract_contact_frequency(freq_2) results in a ContactFrequency where the contact is present in 19/23 frames (~83%). Of course, this only makes sense if you know that the trajectory that makes freq_2 is a subtrajectory of the one that made freq_1.
Those methods were designed specifically for doing this sort of windowing of a contact frequency, so those are the things to use.
Those methods were designed specifically for doing this sort of windowing of a contact frequency, so those are the things to use.
Ah I understand now :) thanks for the explanation