Locations object has no attribute 'dr'

[Fillot]

Describe the bug
Trying to use infer fails.
Code is as follow :

#the localisation (untracked, obtained with trackmate), in x,y,t format
points = pd.read_csv('xyt_coords.txt') 

# here I'm trying to follow the code inside preset_analyser() of the tutorial notebooks.
a = RWAnalyzer()
a.spt_data.from_dataframe(points)
a.spt_data.frame_interval = dt = .01 # in s
a.spt_data.localization_precision = .02 # in µm
a.tesseller = tessellers.KMeans
a.tesseller.ref_distance = 0.02
a.sampler = sampler.Knn(10)
assignment_a = a.sampler.sample(points)

a.mapper = mapper.MapperPlugin('d')
d_map = a.mapper.infer(assignment_a)#<--- this fails here.

This fails with
AttributeError: 'Locations' object has no attribute 'dr'
in this case coming from inference/unsmooth_d.py -> infer_D

I think this is failing because infer_D expects TransLocations, not Locations, which is what it is being fed.
From this, my theory was that I needed to track beforehand, maybe that would put Translocations inside the cells.
I have used

b = RWAnalyzer()
b.spt_data.from_rwa_files('saved2.rwa')
b.spt_data.frame_interval = dt = .01 # in s
b.spt_data.localization_precision = .02 # in µm

b.tracker = tracker.from_non_tracking()
b.tracker.localization_precision = .02
b.tracker.estimated_high_diffusivity = 1
trajs = b.tracker.track('coords_tracked.xyt')

to get a tracking (here trying to load from a rwa file, but this changes nothing anyway.
then (here I was trying to imitate
https://github.com/DecBayComp/Stochastic_Integrals_Diffusivity/blob/master/ito-to-tramway/exemple_8.py

localization_error = 0.01
xyt_file = 'coords_tracked.xyt'
rwa_file = xyt_file.replace('.xyt', '.rwa')

xyt = load_xyt(xyt_file)
analysis_tree = Analyses(xyt)
save_rwa(rwa_file, analysis_tree, force=True)

cells = tessellate(analysis_tree, 'hexagon', ref_distance=0.5,
                           min_location_count=0, label='hexagon')
save_rwa(rwa_file, analysis_tree, force=True)
infer(analysis_tree, 'snr', input_label='hexagon', output_label='snr',
localization_error=localization_error, max_iter=50)

which still fails with the exact same error.

So I'm not sure how exactly I can ensure that infer_D gets cells with translocations attached.
Loading tracked data from a text file, or trying to pass around an Analyses, or loading rwa files doesn't seem to help.

System

• Operating System: Ubuntu 20.04
• Python version: 3.8.6
• TRamWAy version: current (cloned this repo yesterday, 19 nov 2022)

[francoislaurent]

Welcome to this issue tracker, and thanks for asking!

Indeed, all the currently available "plugins" that infer diffusivity require tracked data.

Regarding the second piece of code, I would suggest to run the following snippet instead, similar to
https://tramway-tour.readthedocs.io/en/latest/tracking.html#the-non-tracking-tracker:

b = RWAnalyzer()
b.tracker = tracker.from_non_tracking()
b.tracker.frame_interval = dt = .01 # in s
b.tracker.localization_precision = .02
b.tracker.estimated_high_diffusivity = 1
trajs = b.tracker.track('xyt_coords.txt')

and then, borrowing your first piece of code:

a = RWAnalyzer()
a.spt_data.from_dataframe(trajs)
a.tesseller = tessellers.KMeans
a.tesseller.ref_distance = 0.02
a.sampler = sampler.Knn(10)
assignment_a = a.sampler.sample(trajs)

a.mapper = mapper.MapperPlugin('d')
d_map = a.mapper.infer(assignment_a)

I would expect this to work in this specific order.

Note that, in the first part, I do not feed the RWAnalyzer object with SPT data, as the tracking step will generate them from the localization data.
As a consequence, I do not use the "spt_data" attribute, and instead I initialize the "tracker" attribute and feed into its track method the original localization data file, and not SPT data.

[Fillot]

Thanks, this works!

Am I understanding correctly that the trick is to change/re-initialize RWAnalyser ? I feel like the problem is that feeding untracked data, tracking and assigning will assign the untracked data to the cells (i.e. Locations and not Translocations) which confuses the inference machine.

So I need to feed the analyser tracked data from the start if I want him to have translocations ?

In any case, this works, thank you.