DOC fix random_state in several example for reproducibility#27153
DOC fix random_state in several example for reproducibility#27153glemaitre merged 5 commits intoscikit-learn:mainfrom
Conversation
|
Because the changes are minimal, the visual representations on the images are consistent with how they were before. The only bigger difference is noticeable in the Before 
After 
|
| noisy_moons = datasets.make_moons(n_samples=n_samples, noise=0.05) | ||
| blobs = datasets.make_blobs(n_samples=n_samples, random_state=8) | ||
| no_structure = np.random.rand(n_samples, 2), None | ||
| seed = 170 |
There was a problem hiding this comment.
if it's a number, I'd just use the number in multiple places rather than having a variable (and I'd use 42 to feel better about it 😁 )
There was a problem hiding this comment.
I tried 42 immediately, but it has a radically different visual output then (on the image below). You are right about the number vs. variable, I just package it like this for these seed explorations, to get a similar output as the original as discussed in #26976. I'll revert them when I know the number again. 170 was originally used, so I thought it's better to keep it as the easiest solution. What do you think when looking at the image below compared to the original and the one with 170 as the seed value?
There was a problem hiding this comment.
I see, ok then, we can keep the 170.
There was a problem hiding this comment.
Great, variable removed in 5e4f187, 170 kept as the number. It is a tricky balance with the seed values it seems :)
|
Thanks @TamaraAtanasoska LGTM |
Reference Issues/PRs
Fixes a part of #17568.
What does this implement/fix? Explain your changes.
This PR introduces minor changes in three files:
examples/cluster/plot_linkage_comparison.pyexamples/preprocessing/plot_all_scaling.pyexamples/preprocessing/plot_discretization_classification.pyIn the later two files, just one algorithm in each was missing a
random_stateparameter. The changes are minorAny other comments?
An updated task list of images/files to address is found at the bottom of #17568, see: #17568 (comment). Some files are newly marked as done, but they aren't part of this PR. This is because the
random_statewas already implemented in all the relevant places.@glemaitre @adrinjalali please take a look 👋