jni / i2k-skimage-napari

Repository for the scikit-image, napari, & friends tutorial at I2K 2020

i2k-skimage-napari

Repository for the scikit-image, napari, & friends tutorial at I2K 2020

Installation

with conda

Use:

conda env create -f environment.yml

then

conda activate i2k

with pip

In an environment including pip, use:

pip install -U -r requirements.txt

We recommend that you use Python 3.8 for this tutorial. Both 3.7 and 3.9 should also work but have not been tested.

Datasets

For the dask tutorial, we are going to be using some 3D + t datasets from the Cell Tracking Challenge, specifically:

• the C. elegans developing embryo training dataset (3GB), OR, if that is too large for you to comfortably download,
• the Chinese Hamster Ovarian (CHO) nuclei overexpressing GFP-PCNA training dataset (98MB)

i2k-skimage-napari's Issues

lesson 3 - datasize

we might want to downsample the embryo by default by a factor of 4 to avoid

viewer = napari.view_image(embryo[:, :, ::2, ::2])

as i was running into this issue napari/napari#1507 otherwise!!

Remove neat feature lesson 1

Remove neat feature lesson 1

lesson 2 denoised call

denoised call should probably live under thresholding section not above as it is first used there

filters are "acausal"

filters are looking into the future to define value at the present, see

I know there's all sorts of ways of dealing with this stuff, but it just feels a bit odd to teach this way

Use np.convolve whole time

Can we just use np.convolve the whole time, I don't think the added explanation of correlate adds much and will only serve to confuse (i didn't appreciate this before looking at example). We can add a - sign to the kernel if we want

lesson 4 - dask impossible image

i'd make the impossible image longer timeseries and smaller in x and y to make scrolling through it a better experience

instead of

impossible_image = da.random.random(
    (80_000, 2048, 2048),
    chunks=(1, 1024, 1024),
)

filters is not defined

in around the 9th code cell of 2_segmentation..., filters is not defined yet at edges = filters.scharr(nuclei)

lesson 2 challenge problems

we might want to give more guidance here, we have this material, but it is on max projection not 3D volume ... https://github.com/sofroniewn/napari-training-course/blob/master/lessons/02_interactive_analysis.ipynb

Could try and adapt though ....

lesson 2 adding points with scale

maxima = morphology.local_maxima(transformed)
viewer.add_points(
    np.transpose(np.nonzero(maxima)),
    scale=spacing, # don't forget scale
    name='bad points',
    size=4,
    n_dimensional=True,
)

the above works nicely, not the use of scale, n_dimensional=True all to make 2D rendering look nice too, and the slightly larger size

lesson 3 - on skipping some dask calls

not sure i love this "skipping" first time, will try and see if i have another approach

jupyter-lab missing from environment

i prefer to use juptyer lab over a plain jupyter notebook, but it is missing from the environment, we might want to add

discussion - 0.4 roadmap

should be the napari 0.4 roadmap instead of our 0.4 roadmap otherwise people may be confused

consider using viewer.window.activate()

in, for instance, the magicgui example, the window is shown behind the browser window. Some users might not realize it has done anything at all ... consider using viewer.window.activate() to bring the window to the front:

for example, in filter_interactive_demo:

    ...
    viewer.window.add_dock_widget(widget)
    viewer.window.activate()
    return viewer, widget

lesson 2 cell surface area

Should this be "surface area" in um^2, or volume in um^3?

lesson 3 missing numpy import

missing import numpy as np before

convolution commutative

Actually, it turns out that we can do it in any order (convolution is `associative`), so we can create a filter that combines both the difference and the mean.

I think you want commutative above instead of associative, both are true, but commutativity is the relevant one here

4 discussion topics

Some potential topics

• napari overview
• magicgui
• plugins

io is not defined

in around the 4th code cell of 2_segmentation_and_regionprops, at nuclei = io.imread('../images/cells.tif') ... io is not defined

lesson 2 ndisplay

viewer.ndisplay doesn't do anything, not needed, isn't part of napari API

Should exercises be typeset the same

I like the formatting of the bottom one best

Sklearn regression default - tutorial 1

Should use liblinear

ndi not defined

in 1_image_filters ndi is used before it's defined ... around cell ~20

noisy_change = ndi.correlate(noisy_signal, np.array([-1, 0, 1]))
fig, ax = plt.subplots()
ax.plot(noisy_signal, label='signal')
ax.plot(noisy_change, linestyle='dashed', label='change')
ax.legend(loc='upper left')
ax.margins(0.1)

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-20-0ea399531763> in <module>
----> 1 noisy_change = ndi.correlate(noisy_signal, np.array([-1, 0, 1]))
      2 fig, ax = plt.subplots()
      3 ax.plot(noisy_signal, label='signal')
      4 ax.plot(noisy_change, linestyle='dashed', label='change')
      5 ax.legend(loc='upper left')

NameError: name 'ndi' is not defined

(note, it is later imported import scipy.ndimage as ndi ... but you must also have it in your auto-imports)