297% cpu usage on `DBNBeatTracker -vvv online -j 1`
sloev opened this issue · 3 comments
Expected behaviour
i expected madmom to be able to run DBNBeatTracker -vvv online -j 1 and only use 1 thread, and maybe not 100% cpu for any threads at all.
Actual behaviour
uses 297 % cpu on my macbookpro 2015
Steps needed to reproduce the behaviour
DBNBeatTracker -vvv online -j 1Information about installed software
python 3.8.6
pip list incl madmom==0.17.dev0:
Package Version
---------------------- ------------
aiofiles 0.6.0
aiohttp 3.8.1
aiohttp-socks 0.6.0
aioprocessing 2.0.1
aiosc 0.1.4
aiosignal 1.2.0
anyio 3.6.1
appdirs 1.4.4
arrow 1.2.3
async-timeout 4.0.2
asyncer 0.0.1
attrs 21.2.0
aubio 0.4.9
beautifulsoup4 4.11.1
binaryornot 0.4.4
black 22.8.0
bleach 5.0.1
briefcase 0.3.9
CacheControl 0.12.11
cachy 0.3.0
cairocffi 1.2.0
CairoSVG 2.5.1
certifi 2021.5.30
cffi 1.15.1
chardet 4.0.0
charset-normalizer 2.0.4
cleo 1.0.0a5
click 8.1.3
coloredlogs 15.0.1
commonmark 0.9.1
cookiecutter 2.1.1
crashtest 0.3.1
cssselect2 0.4.1
cycler 0.10.0
Cython 0.29.32
decorator 4.4.2
defusedxml 0.6.0
distlib 0.3.1
dlib 19.21.1
dmgbuild 1.5.2
docutils 0.17.1
drawSvg 1.8.1
ds-store 1.3.0
dulwich 0.20.46
fastapi 0.80.0
filelock 3.0.12
filetype 1.1.0
flatbuffers 2.0.7
frozenlist 1.3.1
future 0.18.2
gdown 4.5.1
gitdb 4.0.9
GitPython 3.1.27
glfw 2.5.5
h11 0.12.0
h2 4.0.0
hpack 4.0.0
html5lib 1.1
humanfriendly 10.0
hyperframe 6.0.1
idna 3.2
ImageHash 4.2.1
imageio 2.9.0
imageio-ffmpeg 0.4.7
importlib-metadata 4.12.0
importlib-resources 5.9.0
imutils 0.5.4
jaraco.classes 3.2.2
Jinja2 3.1.2
jinja2-time 0.2.0
jsonschema 3.2.0
keyring 23.9.1
kiwisolver 1.3.1
llvmlite 0.39.1
lockfile 0.12.2
Logbook 1.5.3
mac-alias 2.2.0
madmom 0.17.dev0
MarkupSafe 2.1.1
matplotlib 3.3.4
matrix-nio 0.18.6
mido 1.2.10
miniaudio 1.52
more-itertools 8.14.0
moviepy 1.0.3
mpmath 1.2.1
msgpack 1.0.4
multidict 5.1.0
mypy-extensions 0.4.3
networkx 2.5
numba 0.56.2
numpy 1.21.6
numpy-ringbuffer 0.2.2
onnxruntime 1.12.1
opencv-python 4.6.0.66
opencv-python-headless 4.6.0.66
packaging 21.3
pathspec 0.9.0
pexpect 4.8.0
Pillow 9.2.0
pip 22.2.2
pkginfo 1.8.3
pkgutil_resolve_name 1.3.10
platformdirs 2.5.2
poetry 1.2.0
poetry-core 1.1.0
poetry-plugin-export 1.0.6
proglog 0.1.10
protobuf 4.21.6
psutil 5.9.2
ptyprocess 0.7.0
PyAudio 0.2.12
pycparser 2.20
pycryptodome 3.10.1
pydantic 1.10.2
pydub 0.25.1
Pygments 2.13.0
pylev 1.4.0
PyMatting 1.1.8
PyOpenGL 3.1.6
pyparsing 2.4.7
pyrsistent 0.18.0
PySocks 1.7.1
python-daemon 2.3.0
python-dateutil 2.8.1
python-magic 0.4.24
python-multipart 0.0.5
python-osc 1.8.0
python-slugify 6.1.2
python-socks 1.2.4
PyWavelets 1.1.1
PyYAML 6.0
readme-renderer 37.1
regex 2021.8.3
rembg 2.0.25
requests 2.26.0
requests-toolbelt 0.9.1
rfc3986 2.0.0
rich 12.5.1
rubicon-objc 0.4.2
scikit-image 0.19.3
scipy 1.7.3
setuptools 59.8.0
shellingham 1.5.0
six 1.15.0
smmap 5.0.0
sniffio 1.3.0
SoundCard 0.4.2
sounddevice 0.4.5
SoundFile 0.10.3.post1
soundmeter 0.1.5
soupsieve 2.3.2.post1
starlette 0.19.1
std-nslog 1.0.1
sympy 1.11.1
syphonpy 0.0.2
text-unidecode 1.3
tifffile 2021.3.5
tinycss2 1.1.0
toga 0.2.15
toga-cocoa 0.3.0.dev36
toga-core 0.3.0.dev36
tomli 2.0.1
tomlkit 0.11.4
tqdm 4.64.0
travertino 0.1.3
twine 4.0.1
typing_extensions 4.3.0
unpaddedbase64 2.1.0
urllib3 1.26.6
uvicorn 0.18.3
virtualenv 20.4.2
watchdog 2.1.9
webencodings 0.5.1
wheel 0.37.1
xattr 0.9.9
yarl 1.6.3
youtube-dl 2021.12.17
zipp 3.8.1
is it possible to run madmom DBNBeatTracker using fewer resources?
thanks :-) have a good autumn day
using python 3.9.0 on same system as mentioned above and same packages results in only 38% cpu
Although you can reduce the number of threads with the -j switch, this does not guarantee that only a single CPU core is used since numpy parallelises stuff quite heavily. Please have a look at https://stackoverflow.com/a/31622299.
You can also disable a couple of neural networks. Per default bagging over 8 networks is performed, I do not expect much lower (detection) performance if only 4 or even 2 are used.