I am trying to install TensorFlow (mac M1).
To do the installation, I first set up a virtual environment and then do the pip installation. However, while this works perfectly fine in my home directory, I can not do it at a different place, such as a folder in my Documents catalog.
The commands I am doing are the following:
/usr/bin/python3 -m venv env
source env/bin/activate
pip install tensorflow-macos
As I said, it works fine in the home catalog, but not in a specific folder.
The error I am getting when doing it in a specific folder:
Building h5py requires pkg-config unless the HDF5 path is explicitly specified
error: pkg-config probably not installed: FileNotFoundError(2, 'No such file or directory')
----------------------------------------
ERROR: Failed building wheel for h5py
I have checked that the python versions are the same, as well as pip. But still no success. I have also tried reinstalling NumPy, PEP515, p5py, and wheel. Also tried uninstalling TensorFlow in the home catalog before installing it somewhere else.
Related
I want to install python packages listed in the requirements file of my github repo. However, I have problems installing those python packages into my conda environment.
First of all, I installed conda with Miniforge3-MacOSX-arm64 which supports the M1 with arm64 architecture. However, some specific python packages like onnxruntime I wasn't able to install, because I encountered error messages like that:
ERROR: Could not find a version that satisfies the requirement onnxruntime
ERROR: No matching distribution found for onnxruntime
I assumed that for those specific python packages there is no support yet for the M1.
Therefore, I pursued another approach. I set the settings of Terminal to "Open with Rosetta". The plan is to install the applications of the intel x86_64 architecture and let Rossetta create the binaries to let run on arm64. Then I uninstalled miniforge for arm64 and installed miniforge for x86_64 named Miniforge3-MacOSX-x86_64. With that setup I was able to install all listed python packages of the requirement file and with pip freeze I can also confirm that they have been installed. However, I am somehow not able to use those python packages. For instance if I want to run pytest I get the following error:
zsh: illegal hardware instruction pytest
I assumed Rossetta takes care of that, that I can use applications for x86_64 also on arm64. But somehow it doesn't work. I tried a lot of different things and am out of ideas.
Does anyone know what the problem is? I would be also thankful for advice and suggestions how to properly set up a python environment on Mac M1.
I had the same problem back in 2days ago, I'm using m1 pro. I was trying to install the python packages only using pip but I got a numbers of errors, then I decided to install with conda.
In my case it worked, here is what I've done so far is:
First Enable the open with rosetta in your zsh.
And then,
# create environment in conda
conda create -n venv python=3.8 # with your python version
# activate
conda activate venv
and visit the conda website to look for the packages:
check packages
For suppose if you are looking for pytest packages then you can search it, and you'll get a result like this, with the available package and channel.
You need to enable that specific channel to get that package with this command:
# config channel
conda config --append channels conda-forge # available channel name
# then install
conda install --yes --file requirements.txt
Make sure, your have the same version of pytest in your requirements.txt file. (eg:pytest==6.2.5)
Hope this should work, if not try to install it with pip like:
pip install -r requirements.txt
after environment enable.
I followed along with the package distribution tutorial for uploading a package on test.pypi.org from a Linux machine, and proceeded to try to install the package on a separate Windows machine.
On the Windows machine, I installed Python 3.8 and ticked the Add Python 3.8 to PATH box. No previous Python installation was present, and it installed to C:\Users\me\AppData\Local\Programs\Python\Python38\.
In Environmental Variables, my PATH under "Users variables" is: C:\Users\me\AppData\Local\Programs\Python38\Scripts\;C:\Users\me\AppData\Local\Programs\Python38\;%USERPROFILE%\AppData\Local\Microsoft\WindowsApps.
My PATH under "System variables" is much longer, but it contains no Python directories.
I then downloaded and installed pip.
In Command Prompt, I first created a virtual environment with py -m venv tutorial and activated it with Scripts\activate. Activation was verified by the (tutorial) next to the prompt. Then, in the virtual env, with py -m pip install ... I installed my example package.
I can verify it's installed by seeing it in the output of py -m pip list:
Package Version
----------------------- -------
example-pkg-me 0.0.1
pip 19.2.3
setuptools 41.2.0
example-pkg-me does not appear outside the virtual env.
The tutorial finishes by having you attempt an import in the Python interpreter:
py
>>> import example_pkg
but I get ModuleNotFoundError: No module named `example_pkg`. Why? Is something wrong with either of my PATHs? Note that the tutorial specifically mentions omitting the username (and import example_pkg_me issues the same error regardless).
Curiously, when trying to install and import a "mainstream (?)" package, it succeeds with no errors. I.e.,
py -m pip install numpy
and
py
>>> import numpy
works.
Update:
Same results on a separate Linux machine with the same steps where the only differences are that Python and pip were already present.
As an aside, if your package is intended to be run (as a GUI app) rather than imported in the interpreter as in the tutorial, how would the person who installed it run it? Would they have to navigate to the directory where the package files were physically installed and run a main.py script there? Is there a better way (like freezing the package as an executable with the likes of cx_Freeze and distributing the EXE instead of relying on pip installations)? Or maybe just expect them to download a repository, run py setup.py install, and then a main.py script?
I'm using anaconda navigator, and had some issues with installing python2 but I eventually got it figured out. However, now every program in anaconda when using python3 looks for packages when importing in anaconda's py36 lib folder, instead of anaconda's lib folder. However, pip is installing into the anaconda lib folder.
I want to change the pip install to the py36 lib folder. However, when I run
pip install --root = "C:/Users/user/AppData/Local/Continuum/anaconda3/envs/py36/lib/site-packages"
I get the error
Directory 'C:/Users/user/AppData/Local/Continuum/anaconda3/envs/py36/lib/site-packages' is not installable. File 'setup.py' not found.
What I don't understand is that the current install location doesn't have a setup.py file in it either.
Does anyone know a solution? Thanks
I'm trying to install a set of libraries used in a data science course, following the instructions from this url, under "Setting up the environment". I have installed Acaconda3 using Python 3.6 on a Windows 10 machine. I have a previously installed version of Python 2.7 which I suspect is the problem, but I´m not sure how to solve this (both versions are now listed in the environment settings).
The problem starts when trying to install the libraries, which gives me the following error when I open a command prompt and enter a "conda install ... " command. The .txt file I´m trying to read is in the project folder. How can I install the packages properly? (anaconda administrator allows a workaround by letting you select the packages and install them from there, but I´d like to know why the command prompt doesn´t work in this case).
(packt-py36) C:\Users\Eric\AnacondaProjects>conda install --file
conda-reqs.txt --yes
Fetching package metadata .............
PackageNotFoundError: Package not found: Conda could not find '
It isn't working because the list of packages was created for OS X. I was able to successfully import these packages on OS X but there are several packages and versions that are not compatible. I have gone through the file and eliminated any non-compatible packages for you. Quite a few packages are missing but you will be able to install from this.
This file may be used to create an environment using:
$ conda create --name <env> --file <this file>
platform: win-64
bleach=1.5.0
cycler=0.10.0=py36_0
decorator=4.0.11=py36_0
entrypoints=0.2.2=py36_1
freetype=2.5.5
html5lib=0.999=py36_0
icu=57.1
ipykernel=4.6.1=py36_0
ipython=6.0.0=py36_0
ipython_genutils=0.2.0=py36_0
ipywidgets=6.0.0=py36_0
jinja2=2.9.6=py36_0
jsonschema=2.6.0=py36_0
jupyter=1.0.0=py36_3
jupyter_client=5.0.1=py36_0
jupyter_console=5.1.0=py36_0
jupyter_core=4.3.0=py36_0
libpng=1.6.27
markupsafe=0.23=py36_2
matplotlib=2.0.0=np112py36_0
mistune=0.7.4=py36_0
mkl=2017.0.1=0
nbconvert=5.1.1=py36_0
nbformat=4.3.0=py36_0
notebook=5.0.0=py36_0
numpy=1.12.1=py36_0
openssl=1.0.2k
pandas=0.19.2=np112py36_1
pandocfilters=1.4.1=py36_0
path.py=10.3.1=py36_0
pickleshare=0.7.4=py36_0
pip=9.0.1=py36_1
prompt_toolkit=1.0.14=py36_0
pygments=2.2.0=py36_0
pyparsing=2.1.4=py36_0
pyqt=5.6.0=py36_2
python=3.6.1=0
python-dateutil=2.6.0=py36_0
pytz=2017.2=py36_0
pyzmq=16.0.2=py36_0
qt=5.6.2
qtconsole=4.3.0=py36_0
scikit-learn=0.18.1=np112py36_1
scipy=0.19.0=np112py36_0
simplegeneric=0.8.1=py36_1
sip=4.18=py36_0
six=1.10.0=py36_0
sqlite=3.13.0
testpath=0.3=py36_0
tk=8.5.18
tornado=4.5.1=py36_0
traitlets=4.3.2=py36_0
wcwidth=0.1.7=py36_0
wheel=0.29.0=py36_0
widgetsnbextension=2.0.0=py36_0
zlib=1.2.8
I'm finding that my pythonpath environment variable is ignored. I'm using python 2.6 on ubuntu. I have in my .bashrc the following:
export PTYHONPATH=/my/home/mylibs/lib/python2.6/site-packages/:$PYTHONPATH
Then I install a new version of numpy using:
python setup.py install --prefix=/my/home/mylibs/
and it gets correctly installed locally. However, when I try to install other packages (also using setup.py) that depend on the new version of numpy, they cannot find it, because by default the loaded numpy is the one in /usr/llib, and not the one specified in my PYTHONPATH. My PYTHONPATH gets correctly set but the system-wide directory is still overruling it.
How can this be fixed? I just want my local version of numpy to be accessed when I do import numpy. I saw other posts related to this with python 2.4 but as far as I can tell it never got resolved. Also, i'd like to do this without installing pip or virtualenv for now. It seems like it should be possible using --prefix or --home options passed to setup.py and then alteration of PYTHONPATH but this does not work for me... the system wide lib dirs are read first.
edit: I try to follow the suggestions and use pip. I have a system wide install of an old pip that does not recognize --user (ver 0.3). I tried to upgrade pip with pip itself but of course that failed because I cannot install it locally, so pip install pip --upgrade --user is not an option. I downloaded a new version of pip and installed locally in my home directory but the system wide old one is still used when I type pip at the prompt. I looked into the pip package and found runner.py so I tried to use it to install packages using:
runner.py install --user numpy --upgrade
That still fails with permission denied:
OSError: [Errno 13] Permission denied: '/usr/bin/f2py2.6'
It looks like --user is broken. I also am not sure how this would solve the fact that the system wide python uses the system wide packages in /usr/lib... is there a solution to this? It seems like it's virtually impossible to install local packages in python nowadays.
Ok, Python will use the first package it finds. The PYTHONPATH gets appended to sys.path, after the system one. So it will normally find the system one first. But the "official" per-user packages directory seems to be placed before that. So create your personal site-packages directory:
mkdir -p $HOME/.local/lib64/python2.7/site-packages
mkdir $HOME/bin
(You may have to change "lib64" to "lib32" or just "lib")
This directory gets placed before the system one on my system. But you should verify it by printing out sys.path.
Then install your packages into there. However, the --user option in the latest pip version should already place it there.
As a list resort you can manipulate sys.path. You can insert your directory into sys.path before the system site-packages, then import numpy.
You are getting permissions errors from the scripts installation, trying to put that in the system location. You can pass additional options to install scripts in your $HOME/bin directory.
Install like this:
pip install --user --install-option="--install-scripts=$HOME/bin"