Ugh great.
I sing mild praises of #Python documentation, but when it comes to third party libraries like #NumPy and especially #SciPy the documentation ranges from less than stellar to utterly atrocious.
And now I have a problem: scipy.optimize.fsolve() is throwing a numpy DeprecationWarning: Conversion of an array with nim > 0 to a scalar is deprecated.
The function I'm passing in to fsolve() returns a scalar!! And I'm already extracting index 0 from the return of fsolve!! WTF???
87/97
Circling back to our program details! Our Beginner Tutorial Track is for absolute beginners, covering the fundamentals: intro to #Python, #NumPy, #Pandas, #ScikitLearn, and #DataVisualization.
Announcing Python-Blosc2 3.6.1
!Unlock new levels of data manipulation with Blosc2! 
We've introduced a major improvement: powerful fancy indexing and orthogonal indexing for Blosc2 arrays.
We've tamed the complexity of fancy indexing to make it intuitive, efficient, and consistent with NumPy's behavior. 
Read all about it on our blog! 
https://www.blosc.org/posts/blosc2-fancy-indexing/
Compress Better, Compute Bigger!
A new article on my site: Approximate first principal component
It describes simple trick to estimate the direction of most spread of a bunch of points without running a full PCA. Includes my Python #NumPy re-implementation and plots of some quick tests.
I move 1 along the x-axis, then rotate by an angle theta, I move 1/phi, rotate by theta, I move 1/phi², rotate by theta etc etc.
Made with #python #numpy #matplotlib
New article on my site: HyAB k-means for color quantization
In which I try to improve color clustering with a different distance function. It's a simple technique in the end but it's pretty hard to evaluate if it's an improvement or not!
https://30fps.net/pages/hyab-kmeans/
https://github.com/pekkavaa/HyAB-kmeans/ with #numpy code
And btw, can I get units for variables already? Annoyed that I can't say that this variable is in inches, or in bytes. Want to be able to tag data that says this is xyY, or XYZ space so that my function can convert if necessary or not.
Great news for OpenMP on Python!
NumPy 2.3 includes early OpenMP support, making sorting operations like np.sort and np.argsort faster by using multiple processor cores — a big step for performance!
This new feature is off by default but can be turned on during installation with -Denable_openmp=true
This marks the beginning of more parallel computing support in NumPy!
If you're writing python libraries, DON'T REQUIRE fileno ON FILE OBJECTS!
Dealing w/ the bullshit that numpy.fromfile wants the fileno attribute on a file object. Yes, it's slightly faster, but it also makes it harder to mock when doing testing.
Now I'm going to have to deal w/ creating a temporary directory, writing the file, and cleaning up afterward. Things that unittest.TestCase should have an option to do, but doesn't. Luckily I've dealt w/ this BS before, so I'll just copy the code from another project.
Since I couldn't figure out how to use numpy.take, and LLMs couldn't figure out how to do what I needed to do, I read the numpy slicing chapter, and I came up with the following:
indexes = np.arange(W * H)
rgb[0, i].flat = c[0].flat[np.array(reps[i].flat) * (W * H) + indexes]
EDIT: it was broken, needed to add in the position index and simplification.
Whoever decided that the plural of "zero" is "zeros" and not "zeroes" has caused me no end of trouble when writing numpy code.
Back to the #numpy & #flocking experiment based on Nicolas Rougier's example in "From Python to Numpy". Code at: https://github.com/villares/sketch-a-day/tree/main/2025/sketch_2025_06_26
More sketch-a-day: https://abav.lugaralgum.com/sketch-a-day
If you like this, support my work:
https://www.paypal.com/donate/?hosted_button_id=5B4MZ78C9J724
https://liberapay.com/Villares
https://wise.com/pay/me/alexandrev562 #Processing #Python #py5 #CreativeCoding
We are pleased to announce the integration of a new stack feature in #Blosc2 , which allows for stacking large arrays along a new axis.
Performance benchmarks show that while aligned chunks yield the best results, #Blosc2 with unaligned chunks can still outperform #NumPy—a welcome discovery! 
Many thanks to Luke Shaw for his excellent work on this new functionality. 
We've updated our recent blog post:
Check it out! 
https://www.blosc.org/posts/blosc2-new-concatenate/#stacking-arrays
Not very good WIP... I always struggle to make #trimesh 3D meshes from scratch with #numpy Code at: https://github.com/villares/sketch-a-day/tree/main/2025/sketch_2025_06_22
More sketch-a-day: https://abav.lugaralgum.com/sketch-a-day
If you like this, support my work:
https://www.paypal.com/donate/?hosted_button_id=5B4MZ78C9J724
https://liberapay.com/Villares
https://wise.com/pay/me/alexandrev562 #Processing #Python #py5 #CreativeCoding
In NumPy, if a and b are matrices, a @ b gives you the matrix product.
If a and b are vectors, a @ b gives you a dot product.
If a and b are scalars (0-dimensional tensors), a @ b throws an error.
Provocative writing style but an interesting proposal for a better array interface for #Python. If you've ever found yourself hacking `None` into #NumPy array indices to make broadcasting work, you might find this (and the associated "I don't like NumPy") interesting:
(via @nicholdav)
This might be more evidence for my growing feeling that JAX is magical.
Today I was scratching my head why k-means didn't seem to reduce Mean Squared Error. The clustering seemed fine. Is the error computation broken? How can it be when it's so simple:
def compute_mse(a, b):
return np.mean((a-b)**2)
Well, arguments 'a' and 'b' were images with 8 bits per channel, so...
Qiita - 人気の記事 