annotate with two different colors based on threshold

[RebeccaWPerry]

I added the capability to annotate an image with features of two different colors. I found this useful for playing around with cutoff thresholds. e.g. see how many features you miss if you increase the minimum size.

[tacaswell]

You should add all new (kw)arguments at the end to maintain back-compatibility. You can call this function with nothing but positional arguments, change change will break the code for anyone who passes 5 or more positional args.

[danielballan]

Thanks for the PR! This seems useful.

What do you think about this usage:

annotate(centroids, image, color=['green', 'purple'], split_category='size', split_thresh=1)

which would also allow usages like

annotate(centroids, image, color=['green', 'purple', 'red'], split_category='size', split_thresh=[1, 10])

where color is expected to be a list with len(split_thresh) - 1 elements. If split_thresh is just a number -- as in the simplest usage -- then color is expected to have two elements.

The main point of my suggestion is passing colors as a list, which is in my opinion more idiomatic and natural than color2. The possibility of passing multiple splits and colors is a bonus.

Let me know what you think. Thanks again.

[RebeccaWPerry]

I agree that the list implementation is a nice way to do it. Would that be backwards compatible though? Maybe we'd have to take special care to handle it if the color passed was a single string rather than a list.

What should the description for the color be that is more generic than "string". I'd be happy to change it to whatever concisely captures all of the ways to format a color. I think "string" is what was there before for color when there was just a single color. I just copied that.

[tacaswell]

I think something like:

from itertools import tee
from six.moves import izip  # I think

# https://docs.python.org/2/library/itertools.html
def pairwise(iterable):
    "s -> (s0,s1), (s1,s2), (s2, s3), ..."
    a, b = tee(iterable)
    next(b, None)
    return izip(a, b)

# change the default of color -> None, clean it up in the code
if color is None:
    color = 'g'

if split_catagory is None:
    # old behavior, just assume color is well behaved

else:
    if len(color) != len(slpit_thresholds) + 1:
        raise ValueError("something clever about length miss-match")
    low = np.where(centroids[split_category] < split_thresh[0])[0]
    ax.scatter(centroids['x'][low], centroids['y'][low], 
                   s=circle_size, facecolors='none', edgecolors=color[0])

    for c, (bot, top) in izip(color[1:-1], pairwise(split_thresholds)):
        # not sure we strictly need the where, you can pass bool arrays
        # directly to np array, not sure if pandas plays nice (probably not)
        indx = np.where((centroids[split_category]) >= bot *
                        (centroids[split_category]) < top)[0]
            ax.scatter(centroids['x'][indx], centroids['y'][indx], 
                   s=circle_size, facecolors='none', edgecolors=c)

    high = np.where(centroids[split_category] >= split_thresholds[-1])[0]
    ax.scatter(centroids['x'][low], centroids['y'][low], 
                   s=circle_size, facecolors='none', edgecolors=color[-1])

Would do this for an arbitrary number of splits and be back-compatible. You get some strange naming (color can be a list of colors), but it works.

[RebeccaWPerry]

Thanks for the feedback. I updated the code in the pull request to allow for splitting into an arbitrary number of sections, got rid of the np.where()'s, and made it backwards compatible. I tested it with these examples:

import trackpy as tp

frames = tp.ImageSequence('Desktop/desktop/copiedData2/22/image0001.tif')

f = tp.locate(frames[0],3)

# check successful
tp.annotate(f, frames[0])
tp.annotate(f, frames[0],color='purple')
tp.annotate(f, frames[0],color=['purple','green','red'],split_category='signal',split_thresh=[10,20])
tp.annotate(f, frames[0],color=['purple','green','red'],split_category='mass',split_thresh=[110,120])


# check exceptions:
tp.annotate(f, frames[0], color=[[.1,.1,.5],[.1,.8,0.],[1,0,0]])
tp.annotate(f, frames[0], color=[[.1,.1,.5],[.1,.8,0.],[1,0,0]], split_category = 'signal', split_thresh=10)

[tacaswell]

should be split_thresh is not None and ... I think

[RebeccaWPerry]

This is to handle the case where someone puts a single number in as a threshold rather than a list. It must be converted to a list because later split_thresh[0] needs to return a value. Possible split_thresh: [10], 10, 10.0, [10,12,13], etc.

[nkeim]

Some people (like me, sometimes) have the habit of using tuples for sequence literals. isinstance(split_thresh, (list, tuple)) would save those folks from seeing a confusing exception.

I know I haven't said much on this PR, but I'm looking forward to this being in trackpy.

[nkeim]

I know the train's already left the station, but I don't see how this test wouldn't always evaluate to True. Did you mean to write if not isinstance(split_thresh, Iterable)?

In any case, a string also counts as iterable, so there has to be another way.

Sorry I've been doing nothing but writing comments. I'll be back on my development computer by tomorrow, I promise :).

[danielballan]

My fault. This is what I get for late-night merging.

On Wednesday, July 9, 2014, Nathan Keim notifications@github.com wrote:

In trackpy/plots.py:

 Returns
 ------
 axes
 """
 import matplotlib.pyplot as plt

• from itertools import tee, izip
• from collections import Iterable

• https://docs.python.org/2/library/itertools.html

• def pairwise(iterable):

•    "s -> (s0,s1), (s1,s2), (s2, s3), ..."
  

•    a, b = tee(iterable)
  

•    next(b, None)
  

•    return izip(a, b)
  
• if color is None:

•    color = 'g'
  

• if not (split_thresh, Iterable):

I know the train's already left the station, but I don't see how this test
wouldn't always evaluate to True. Did you mean to write if not
isinstance(split_thresh, Iterable)?

In any case, a string also counts as iterable, so there has to be another
way.

Sorry I've been doing nothing but writing comments. I'll be back on my
development computer by tomorrow, I promise :).

—
Reply to this email directly or view it on GitHub
https://github.com/soft-matter/trackpy/pull/104/files#r14724711.

[RebeccaWPerry]

This bit of code is to handle the case where split_thresh is a single number. I turn it into a list because split_thresh[0] needs to be valid

[nkeim]

I think that the code, as written, will never put anything into a list, scalar or not. A multi-element tuple always evaluates to True.

I can say that I definitely get the wrong behavior when I try this code at the IPython prompt.

[RebeccaWPerry]

This is the case I was trying to handle:
tp.annotate(f, frames[0],color=['purple','green'],split_category='signal',split_thresh=10)
I see now that it is broken. What is the proper way to check if something is iterable?

[tacaswell]

isinstance(foo, Iterable) will return True if foo is iterable (and properly implements the ABC stuff from collections).

[nkeim]

Thanks, Tom! I've been doing it a much less elegant way. I guess the only complication is if you'd like strings to count as single values; then you'd need something like isinstance(foo, Iterable) and not isinstance(foo, str). But it looks like that's not the case in this code.

[tacaswell]

A wonderful phrase I learned last night, 'duck typing brings a can of worms: either you go fishing with them or eat them'

[RebeccaWPerry]

I made this change in my local branch. How should I go about submitting this one little fix to the main repository? A new pull request?

[tacaswell]

A new PR is best, it is only easy to add commits to a PR before it has been merged.