Scientific Computing

The unofficial OpenCV PyPI wheels work with pip install methods:

pip install opencv-python

For ARM / Raspberry Pi:

pip install opencv-python

also works for certain ARM platforms like the Raspberry Pi.

OpenCV is trivial and fast to install on a Raspberry Pi via pip as described above.

For the latest extended functionality that hasn’t yet been incorporated into the core package, OpenCV including the Extra contributed modules may be obtained by:

pip install opencv-contrib-python

• OpenCV release notes
• Conda-forge OpenCV 3
• (reference) Windows-only OpenCV wheels. (cpMN where you have Python M.N). contrib includes OpenCV-extra packages.

If ... is not supported on this platform error be sure it’s using desired Python install. One may have to manually specify the path for the pip command e.g.

$Env:SystemDrive\anaconda3\scripts\python -m pip install opencv-python

We made several test scripts to try out the OpenCV install. Compiling OpenCV yourself allows customizing and optimizing OpenCV for your computer (e.g. using GPU, TBB, OpenCL, etc.).

The conda install opencv and conda install -c conda-forge opencv methods for OpenCV continue to be BROKEN for video/image reading and display. Use pip install above instead.

If YouTube won’t accept a video upload, or the video never completes “Processing” on YouTube, try re-encoding the video with FFmpeg. Sometimes lossy conversion is necessary to achieve the YouTube recommended upload settings.

Periscope video downloads use MPEG TS container with .ts file extension. These .ts files can be played back in VLC or similar to confirm content. A lossless conversion to YouTube is possible with:

ffmpeg -i pscp.ts -bsf:a aac_adtstoasc -codec copy -max_muxing_queue_size 1000 pscp.mp4

The FFmpeg option -max_muxing_queue_size is an arbitrary parameter. The default queue size is sometimes not large enough:

Too many packets buffered for output stream 0:1.

• Do not use .mkv extension for YouTube uploads, it will fail to process.
• certain formats like AV1 or FFV1 will fail to finish processing the upload
• after completing Uploading, video should start Processing in less than 5 minutes. If not, it will probably never actually work.

Uploads that don’t finish conversion processing after upload are not viewable on YouTube.

Generating a range of datetime data is a common data analysis and simulation task. Here we show examples of generating datetime vectors for Python datetime and numpy.datetime64

Python datetime

Python datetime is often used as timezone-naïve with UTC as the assumed timezone. This custom avoids ambiguities when working with Pandas and Numpy, which are foundational for Python data science.

Generate a range of Python datetime like:

from __future__ import annotation
from datetime import datetime, timedelta


def datetime_range(start: datetime, end: datetime, step: timedelta) -> list[datetime]:
    """like range() for datetime"""
    return [start + i * step for i in range((end - start) // step)]


dt = datetime_range(datetime(2019, 12, 1), datetime(2020, 4, 1), timedelta(days=1))

Numpy datetime64

Numpy datetime64 generates a range of times like:

dt = numpy.arange('2019-12-01', '2020-04-01', dtype='datetime64[D]')

Pandas has the date_range function to generate time vectors.

Matlab datetime works much like Python datetime, and is generally recommended. Matlab plotting functions generally support “datetime” class. It is sometimes necessary to manipulate Matlab plots involving datetime for the desired result.

dt = datetime(2019, 12, 1):datetime(2020, 3, 1);

dat = rand([100, length(dt)]);
y = 1:size(dat, 1);

pcolor(dt, y, dat)

If the plot doesn’t have the desired datetime axis formatting, try datetick.

datetick('x', 'yyyy-mm-dd', 'keepticks')

Generate vector of datetime. Python matplotlib supports datetime.datetime and numpy.datetime64 in most plot functions.

After updating an operating system graphics driver while Matlab was open, an error may occur upon plotting in Matlab like:

MATLAB has experienced a low-level graphics error, and may not have drawn correctly. Read about what you can do to prevent this issue at Resolving Low-Level Graphics Issues then restart MATLAB.

The solution to this issue typically is, as the message suggests, just restarting Matlab. A full reboot is not always necessary, but try that if the error persists.

Skipping a test with Meson or CMake by returning error code 77 is a de facto practice. Sometimes it is only feasible to know a test should be skipped by attempting to run that test.

Meson build system accepts return code 77 by default as a signal to skip the test. Configure CMake to skip return code:

add_test(NAME MyTest COMMAND mytest)  # arbitrary

set_property(TEST MyTest PROPERTY SKIP_RETURN_CODE 77)

The vcgencmd utility allows reading a few dozen hardware measurements on the Raspberry Pi boards.

CPU Temperature is checked by:

vcgencmd measure_temp

Typical temperatures in office environment, with case:

A red thermometer icon GPU-superimposed on the Raspberry Pi display output means the Raspberry Pi is overheating and is throttling the CPU and GPU to avoid self-destruction.

Raspberry Pi 0, 1, 2, 3 temperature thresholds:

CPU temp. [C]	icon	throttle
< 80	none	none
80 - 85		CPU
> 85		CPU & GPU

Raspberry Pi 4 temperature thresholds:

CPU temp. [C]	icon	throttle
80 - 85	none	CPU: 1000 MHz
85 - 90		CPU
> 90		CPU & GPU

The current Raspberry Pi CPU clock speed is obtained from

vcgencmd measure_clock arm

The output is in units of Hertz. The Raspberry Pi CPU clock speed and power consumption is dynamic:

Log temperature measurements with crontab -e. This can periodically log temperature and CPU frequency, e.g. add a line like:

@hourly  vcgencmd measure_temp | /usr/bin/logger

logger writes the measured parameters into the system log.

Related: Monitor Raspberry Pi DC input voltage

Many basic Git use cases involve a main branch with feature branches periodically merged into the main branch. For certain purposes, totally distinct branch without a common history can exist in the same Git repo. One of the most common uses of this is for documentation. For example, GitHub will build a website from the gh-pages branch.

Setup blank Git branch

Do NOT force push during this procedure, you may accidentally erase years of work!

This example assumes you want to create a gh-pages empty branch for documentation on GitHub, but will of course work for other purposes too.

From the repo directory create a blank Git branch:

git switch --orphan gh-pages

Remove the unneeded files from this branch (by default, all existing files are staged from the previous branch)

git rm --cached -r .

git clean -id

This leaves the .git/ directory, which should not be disturbed.

Copy documentation files

What happens next depends on if your documentation files were already added to another branch (tracked) or were not added to Git (untracked). Assume wanted files for the blank gh-pages branch are in docs/ on main branch.

copy over the files to gh-pages

git checkout main -- docs/

git commit -am "moved docs"

Upload files

1. push the documentation

   git push -u origin gh-pages

2. Enable the documentation builds from github.invalid/username/myrepo/settings → GitHub Pages under Source select gh-pages.

3. In a few minutes, the webpages at username.github.io/myrepo/ should be active.

Once everything is working, the old docs/ folder isn’t needed.

Python 3.6 f-strings have been shown to be the fastest string formatting method in microbenchmarks by Python core dev Raymond Hettinger benchmark relative speed factors:

• f-string: 1.0
• concatenate string: 1.33
• join sequence of strings: 1.73
• %s formatting operator: 2.40
• .format() method: 3.62
• Template() method: 11.48

The reason for this speedy performance was described by Python core dev Serhiy Storchaka.

Goldwave 5.x is stable and the recommended version to use on Linux via WINE. An open source alternative to Goldwave is Audacity. If using Goldwave on Windows directly:

winget install --id=GoldWave.GoldWave

To use Goldwave on Linux via WINE:

Install Goldwave prereqs. 32-bit WINEPREFIX is mandatory due to wmp10 prereq.

WINEPREFIX=~/.wine32 WINEARCH=win32 winetricks wmp10

This creates a 32-bit winearch (default .wine directory is 64-bit). The default install options are fine.

Install Goldwave in WINE.

WINEPREFIX=~/.wine32 wine gwave5*.exe

This also creates a Goldwave icon in the Ubuntu apps menu.

Configure soundcard in Goldwave by:

1. F11 key → Control Properties → System tab
2. click Use DirectSound AP radio button
3. click OK to save configuration

Select “loopback” what you hear audio under F11 → Device → Record and pick Loopback Pulseaudio.

Test Goldwave 5 by F11 key → Device tab and click Test playback button to hear a brief test tone. Create new/save/play sound files with Goldwave 5.7 in Linux WINE

These settings are useful for Goldwave on Windows as well as WINE/Linux.

• Constantly monitor audio levels: F11 → Record → monitor input
• label VU graph axes in dB: Right click VU meter → Properties → Show Axis

It’s recommended to use Goldwave 5.x as above on WINE instead. There are some errors but Goldwave 6 still works to record and playback on WINE ≥ 2 using a Windows 7 Wineprefix.

1. Download Goldwave 6
2. configure Wineprefix for Windows 7 using winecfg → Applications → Windows Version → Windows 7
3. Install Goldwave 6 on WINE

wine InstallGoldwave6*.exe

As of WINE 2.1, error upon opening Goldwave 6 include

Floating point overflow

and there are also

Unhandled exception

on closing Goldwave 6 and it will keep asking to reset Goldwave settings. Nonetheless Goldwave 6 does record/play on WINE 2.1.