Remove rng parameter of binary_blobs() function in plot_random_walker_segmentation.py

[aymuos15]

Using the rng parameter throws the following TypeError:

TypeError: binary_blobs() got an unexpected keyword argument 'rng'

Description

Parameter rng removal in line:

data = skimage.img_as_float(binary_blobs(length=128, rng=1))

of Random walker segmentation Example.

## Removed rng parameter.
data = skimage.img_as_float(binary_blobs(length=128))

To reproduce error and fix: https://colab.research.google.com/drive/1iy8InDNAoadAaNkn0WuyND6S0a3VFmEx?usp=sharing

Removing it does not make any change to the aim of the tutorial; which is showing the segmentation capabilities.

[lagru]

Hey @aymuos15, thanks for the suggestion. I am a bit confused why you get the TypeError. rng is a new argument that replaces the deprecated seed. Google doesn't allow me to check without sign-in, but I am assuming that your Colab notebook uses an outdated scikit-image version. So for that reason alone I wouldn't remove the parameter.

Removing it does not make any change to the aim of the tutorial; which is showing the segmentation capabilities.

I agree, though I think in general we try to pin our examples so they are completely reproduce-able. So I am hesitant to remove the rng parameter.

[soupault]

Thank you for the PR, @aymuos15 ! I personally do not have a strong preference on what is better here - higher simplicity or higher reproducibility.

[aymuos15]

@lagru You are right. It was a version error. I apologise for not checking before. Was a mistake on my part. Will close the request if that is okay with you. Thank you for pointing it out.

@soupault I personally feel the lesser the parameters in a base example, the better. But this is just a update issue on colab so doesn't matter that much I guess?

[lagru]

Ideally we could have both. It seems we don't use explicit seeds in a lot of gallery examples.

$ grep "default_rng" -r doc/examples/
doc/examples/filters/plot_inpaint.py:rstate = np.random.default_rng(0)
doc/examples/filters/plot_entropy.py:rng = np.random.default_rng()
doc/examples/filters/plot_deconvolution.py:rng = np.random.default_rng()
doc/examples/filters/plot_restoration.py:rng = np.random.default_rng()
doc/examples/features_detection/plot_shape_index.py:    rng = np.random.default_rng()
doc/examples/applications/plot_colocalization_metrics.py:rng = np.random.default_rng()
doc/examples/applications/plot_rank_filters.py:rng = np.random.default_rng()
doc/examples/registration/plot_masked_register_translation.py:rng = np.random.default_rng()
doc/examples/transform/plot_ransac.py:rng = np.random.default_rng()
doc/examples/transform/plot_ssim.py:rng = np.random.default_rng()
doc/examples/transform/plot_fundamental_matrix.py:rng = np.random.default_rng(random_seed)
doc/examples/segmentation/plot_random_walker_segmentation.py:rng = np.random.default_rng()

Most cases are for noise though. But there is precedent for both options.

In this case, I feel like for functions whose API involves randomness it's actually not a bad thing to make its usage visible. So I'm +0.1 on keeping it the way it is.

@aymuos15, feel welcome to leave this open for a few days in case other devs feel different. :)

[stefanv]

In the scipy lecture notes, we use a long integer seed, and explain somewhere why we set a seed in examples.

[aymuos15]

Thank you @stefanv @lagru for the explanation and information. I will keep it open the for the time being.

[mkcor]

I'm in favour of

• either closing this PR,
• or editing it so we follow what @stefanv wrote about the scipy lecture notes:

we use a long integer seed, and explain somewhere why we set a seed in examples.

[lagru]

@aymuos15, do you want to update the PR or make a new one according @mkcor's suggestion? :)

[aymuos15]

Hi all. Apologies for the late response.

I will be in favour of: "we use a long integer seed, and explain somewhere why we set a seed in examples."

Will edit this PR by the end of next week, I am away at the moment. Thank you.