LLVMLite / numba¶
llvmlite cause known installation issues when installing via
There are three circumstances where these errors arise:
1) llvmlite and numba are already installed
In this case, the installation of autolens raises an exception like the one below:
Cannot uninstall 'llvmlite'. It is a distutils installed project and thus we cannot accurately determine which files belong to it which would lead to only a partial uninstall.
This means that
numba are already installed, which you can check as follows:
pip show llvmlite pip show numba
PyAutoLens works fine across many versions of lvvmlite and numba, so you should be ok to circumvent this error by simply not reinstalling these libraries when you install PyAutoLens:
pip install autolens --ignore-installed llvmlite numba
2) llvmlite and numba are not already installed
In this case, a dependency error will arise where one of these libraries could not be installed. If you are trying to
install via pip, we recommend you instead follow the installation via conda instructions
which install these libraries as part of the
A common error for installing llvmlite is that a config file is missing:
Failed to install - No such file or directory: 'llvm-config': 'llvm-config'
The first solution to try is to upgrade your pip via one of the following commands:
pip install --upgrade pip pip3 install --upgrade pip
You may then retry the autolens installation:
pip install autolens
In the above solution fails, you can manually install the following versions
numpy==1.19.7 which are known to work with PyAutoLens:
pip install llvmlite==0.32.1 pip install numba==0.47.0 -ignore-installed llvmlite pip install numpy==1.19.7 pip install autolens --ignore-installed llvmlite numba numpy
This may raise warnings, but PyAutoLens has been tested with this combination of versions which have had less installation issues.
3) The version of numba and numpy clash
If numba and numpy are not on versions compatible with one another the following error can arise when running autolens:
TypeError: expected dtype object, got 'numpy.dtype[float64]'
The easiest solution is to downgrade to
pip install numpy==1.19.7
If you are still facing installation issues please raise an issue on the GitHub issues page.
Current Working Directory¶
PyAutoLens scripts assume that the
autolens_workspace directory is the Python working directory. This means
that, when you run an example script, you should run it from the
autolens_workspace as follows:
cd path/to/autolens_workspace (if you are not already in the autolens_workspace). python3 examples/model/beginner/mass_total__source_parametric.py
The reasons for this are so that PyAutoLens can:
- Load configuration settings from config files in the
- Load example data from the
- Output the results of models fits to your hard-disk to the
- Import modules from the
autolens_workspace, for example
from autolens_workspace.transdimensional import pipelines.
If you have any errors relating to importing modules, loading data or outputting results it is likely because you
are not running the script with the
autolens_workspace as the working directory!
Matplotlib uses the default backend on your computer, as set in the config file:
If unchanged, the backend is set to ‘default’, meaning it will use the backend automatically set up for Python on your system.
[general] backend = default
There have been reports that using the default backend causes crashes when running the test script below (either the code crashes without a error or your computer restarts). If this happens, change the config’s backend until the test works (TKAgg has worked on Linux machines, Qt5Agg has worked on new MACs). For example:
[general] backend = TKAgg