Incoming links: NumPy
XML file format (but with 99% of the action of interest in a domain-specific language on the
CsInstruments and CsScore elements) that can be played and the reference implementation. Offers complex effects out-of-box apparently.Allows you to easily define instruments with seemingly arbitrary mathematical functions, and then use them to play notes at given time intervals.
The instrument functions can be parametrized, and each note played can have different parameters.
The instrument definition actually defines a block diagram graph, much like a hardware synthesizer would.
Csound is so not-bloated that it contains an UI system. And it includes an interactive virtual MIDI keyboard that interacts with parameter knobs: www.csounds.com/manual/html/MidiTop.html
But hey, it's fun. And like any other good domain-specific language, debugging it is barbaric of course.
If only it had been written in Python... the array manipulation boilerplate would be likely perfect for NumPy, and this would have been exactly what Ciro Santilli wanted!
CSound states that one of its design goals is backward compatibility, and it shows. Some of the stuff is utterly arcane, e.g. you have to remember what
GEN10, GEN11, etc. mean instead of having named enums.It just worked on Ubuntu 20.04 no questions asked:which runs this file: github.com/csound/csound/blob/92409ecce053d707360a5794f5f4f6bf5ebf5d24/examples/xanadu.csd and this plays a relly cool sound demo:
sudo apt install csound
git clone https://github.com/csound/csound
cd csound
git checkout 92409ecce053d707360a5794f5f4f6bf5ebf5d24
csound examples/xanadu.csdSave to file instead of playing:or direct ogg output:or pipe to stdout to FFmpeg TODO: stackoverflow.com/questions/64970503/how-to-pipe-csound-output-to-ffmpeg-for-conversion-without-an-intermediate-file
csound -o xanadu.wav xanadu.csdcsound --ogg -o xanadu.ogg xanadu.csdTODO find the most amazing set of songs made with it on GitHub? Some examples:
- www.csounds.com/toots/index.html has a good 101 on instrument design
- Csound FLOSS manual
- iainmccurdy.org/csound.html about 100 CC BY-SA examples. Each is a minimal study showing a specific technique, not a full composition, some seem advanced. Dude's a beast.
- github.com/csound/csound/tree/f2e70825fb543a6b15011c6984371f61ab2a00dd/tests/soak in-tree minimal examples
- github.com/csound/manual/tree/4049b286493d972ff7248b5596e47e7ae97a0cf9/examples contains the examples for the manual which is rendered at: It's insane, but it's fun! Ah those newbs who separate manuals from main tree.
- linuxsynths.com/CsoundPatchesDemos/csound.html on LinuxSynths
- github.com/csound/examples/tree/ae578159328178142c1055c7f78e28b42eb29774/csd as a few dozen examples
- freaknet.org/martin/audio/csound/ 10 pieces with source
Documentation-wise, it's a bit lacking. The only dude who can explain it really well, Dr Richard Boulanger, made the "The Csound Book" closed source, so, congrats, this will forever hurt the popularity of Csound.
Examples:
- csound/sine.csd
- csound/amplitude_frequency.csd
- csound/linen.csd: simple attack/release envelope, documented at: www.csounds.com/manual/html/linen.html
- csound/chorus.csd: chorus effect
- csound/bend.csd: bend using
linseg - csound/vibrato.csd
- csound/crossfade_generators.csd
- csound/table.csd
- csound/virtual_keyboard.csd
See sections: "Example 1 - N even", "Example 2 - N odd" and "Representation in terms of sines and cosines" of www.statlect.com/matrix-algebra/discrete-Fourier-transform-of-a-real-signal
The transform still has complex numbers.
Summary:Therefore, we only need about half of to represent the signal, as the other half can be derived by conjugation.
- is real
"Representation in terms of sines and cosines" from www.statlect.com/matrix-algebra/discrete-Fourier-transform-of-a-real-signal then gives explicit formulas in terms of .
NumPy for example has "Real FFTs" for this: numpy.org/doc/1.24/reference/routines.fft.html#real-ffts
DFT of with 25 points
. Source at: numpy/fft_plot.py. This plot illustrates how the DFT of a real signal is symmetric around the middle point, and so only half of the transform points are needed to reconstruct the original signal. We also see how the phase of the sinusoids determines if their DFT components are real or imaginary.Let's look into a sample plot,
out/manual/plotOut/svg_plots/massFractionSummary.svg, and try to understand as much as we can about what it means and how it was generated.This plot contains how much of each type of mass is present in all cells. Since we simulated just one cell, it will be the same as the results for that cell.
We can see that all of them grow more or less linearly, perhaps as the start of an exponential. We can see that all of them grow more or less linearly, perhaps as the start of an exponential. We can see that all of them grow more or less linearly, perhaps as the start of an exponential.which must correspond to the different
By grepping the title "Cell mass fractions" in the source code, we see the files:
models/ecoli/analysis/cohort/massFractionSummary.py
models/ecoli/analysis/multigen/massFractionSummary.py
models/ecoli/analysis/variant/massFractionSummary.pymassFractionSummary plots throughout different levels of the hierarchy.By reading
models/ecoli/analysis/variant/massFractionSummary.py a little bit, we see that:- the plotting is done with Matplotlib, hurray
- it is reading its data from files under
./out/manual/wildtype_000000/000000/generation_000000/000000/simOut/Mass/, more precisely./out/manual/wildtype_000000/000000/generation_000000/000000/simOut/Mass/columns/<column-name>/data. They are binary files however.Looking at the source forwholecell/io/tablereader.pyshows that those are just a standard NumPy serialization mechanism. Maybe they should have used the Hierarchical Data Format instead.We can also take this opportunity to try and find where the data is coming from.Massfrom the./out/manual/wildtype_000000/000000/generation_000000/000000/simOut/Mass/looks like an ID, so wegrepthat and we reachmodels/ecoli/listeners/mass.py.From this we understand that all data that is to be saved from a simulation must be coming from listeners: likely nothing, or not much, is dumped by default, because otherwise it would take up too much disk space. You have to explicitly say what it is that you want to save via a listener that acts on each time step.
Minimal condition mass fraction plot
. Source. File name: 
out/manual/plotOut/svg_plots/massFractionSummary.svgMore plot types will be explored at time series run variant, where we will contrast two runs with different growth mediums.
Run output is placed under
out/:Some of the output data is stored as
.cpickle files. To observe those files, you need the original Python classes, and therefore you have to be inside Docker, from the host it won't work.We can list all the plots that have been produced under Plots are also available in SVG and PDF formats, e.g.:
out/ withfind -name '*.png'The output directory has a hierarchical structure of type:where:
./out/manual/wildtype_000000/000000/generation_000000/000000/wildtype_000000: variant conditions.wildtypeis a human readable label, and000000is an index amongst the possiblewildtypeconditions. For example, we can have different simulations with different nutrients, or different DNA sequences. An example of this is shown at run variants.000000: initial random seed for the initial cell, likely fed to NumPy'snp.random.seedgenereation_000000: this will increase with generations if we simulate multiple cells, which is supported by the model000000: this will presumably contain the cell index within a generation
We also understand that some of the top level directories contain summaries over all cells, e.g. the
massFractionSummary.pdf plot exists at several levels of the hierarchy:./out/manual/plotOut/massFractionSummary.pdf
./out/manual/wildtype_000000/plotOut/massFractionSummary.pdf
./out/manual/wildtype_000000/000000/plotOut/massFractionSummary.pdf
./out/manual/wildtype_000000/000000/generation_000000/000000/plotOut/massFractionSummary.pdfEach of thoes four levels of
plotOut is generated by a different one of the analysis scripts:./out/manual/plotOut: generated bypython runscripts/manual/analysisVariant.py. Contains comparisons of different variant conditions. We confirm this by looking at the results of run variants../out/manual/wildtype_000000/plotOut: generated bypython runscripts/manual/analysisCohort.py --variant_index 0. TODO not sure how to differentiate between two different labels e.g.wildtype_000000andsomethingElse_000000. If-vis not given, a it just picks the first one alphabetically. TODO not sure how to automatically generate all of those plots without inspecting the directories../out/manual/wildtype_000000/000000/plotOut: generated bypython runscripts/manual/analysisMultigen.py --variant_index 0 --seed 0./out/manual/wildtype_000000/000000/generation_000000/000000/plotOut: generated bypython runscripts/manual/analysisSingle.py --variant_index 0 --seed 0 --generation 0 --daughter 0. Contains information about a single specific cell.
@cirosantilli/_file/python/pytorch/python/pytorch/matmul.py Updated 2024-12-15 +Created 1970-01-01
Matrix multiplication example.
Fundamental since deep learning is mostly matrix multiplication.
NumPy does not automatically use the GPU for it: stackoverflow.com/questions/49605231/does-numpy-automatically-detect-and-use-gpu, and PyTorch is one of the most notable compatible implementations, as it uses the same memory structure as NumPy arrays.
All those dedicated applied mathematicians languages are a waste of society's time, Ciro Santilli sure applied mathematicians are capable of writing a few extra braces in exchange for a sane general purpose language, we should instead just invest in good libraries with fast C bindings for those languages like NumPy where needed, and powerful mainlined integrated development environments.
And when Ciro Santilli see the closed source ones like MATLAB being used, it makes him lose all hope on humanity. Why. As of 2020. Why? In the 1980s, maybe. But in the 2020s?
TODO why is it so hard to find anything non perturbative :-(
- www.youtube.com/channel/UCPHFUHiwbpMqC8ONxEICCiQ NanoNebula using raw Perl PDFL en.wikipedia.org/wiki/Perl_Data_Language (the Perl NumPy)
- www.youtube.com/watch?v=9TJe1Pr5c9Q "Interplay of Quantum Electrodynamics and Quantum Chromodynamics in the Nontrivial Vacuum" by CSSM Visualisation (2019)
On a quantum computer...:
- www.cornell.edu/video/john-preskill-simulating-quantum-field-theory-with-quantum-computer Simulating Quantum Field Theory with a Quantum Computer by John Preskill (2019)
- www.youtube.com/watch?v=Lln-C21u0U8 Quantum Simulation from Quantum Chemistry to Quantum Field Theory by Peter Love (2019)
Are we living in the matrix? by David Tong (2020)
Source. Talks about how the Nielsen-Ninomiya theorem means it is impossible to simulate QFT on a computer in the case of a lattice gauge theory.