How to get full path of resource file from within python script? - python

Ok, so i made a python script places inside a package. The tree looks something like this:
├── foo
│   ├── __init__.py
│   ├── funcs
│   │   ├── __init__.py
│   │   └── stuff.py
│   ├── resources
│   │   └── haarcascade_frontalface_default.xml
│   └── scripts
│      ├── __init__.py
│      └── script.py
└── setup.py
So inside the script file, im using openCV's cv2 to detect faces, and for that the cv2.CascadeClassifier requires the path of the XML file located under /resources. Now because this is a script, i need to be able to run it from anywhere, so a relative path to the resource file sadly doesn't do the trick. How can I get the absolute path to the xml file from within script.py? You can assume that the script and the xml file is located relative to each other respectively, just like the example above. Thanks :))
PS: Bonus if the solution works with eggs as well. Much appreciated

Currently the best way to do this is importlib.resources. Since Python 3.7 it is available in the standard library. For earlier versions, there is a backport called importlib_resources.
Follow the documentation.
In your case this should more or less look like this :
import importlib.resources
xml_path = importlib.resources.path('foo.resources', 'haarcascade_frontalface_default.xml')
There are many advantages to this, most importantly it is standard and it will work wherever the package is installed even if it's in a zip file.
In your case, you might have to add an __init__.py file to your resources directory.

Using the os module works, but if you have access to a python version >= 3.4, then pathlib is an alternative that handles itself a little easier and performs better across platforms:
from pathlib import Path
# when using pathlib.Path, slashes get automatically transformed into the
# correct path-division character, depending on the platform
RESOURCES_PATH = Path(__file__).parent.parent / "resources"
face_cascade = cv2.CascadeClassifier()
face_cascade.load(RESOURCES_PATH / "haarcascade_frontalface_default.xml")
If you find yourself defining lots of these kinds of constants, consider putting all of them in a file like foo/util.py so that they are easily reusable within your project and don't need to be re-declared or imported from a script.
An even better option in python versions >=3.7 is using importlib.resources.path, which resolves resources automatically from the package root, so you don't need to find it by hand by walking up from __file__:
import importlib
face_cascade = cv2.CascadeClassifier()
with importlib.resources.path("foo.resources", "haarcascade_frontalface_default.xml") as haar_resource:
# haar_resource is a pathlib.Path object here as well, so plugging it is simple
face_cascade.load(haar_resource)
This is a lot more elegant and should be the preferred solution given it's available.

I'm not sure I understand the question correctly, but maybe os.path will help? Something like:
>>> import os
>>> os.path.abspath("directory/somefile.txt")
'C:/somedirectory/directory/directory/somefile.txt'

Related

Python packages/modules in different subdirs

I'm relatively new to Python, and I'm struggling to understand if what I'm trying to do is just not possible, or if I am just doing it wrong. I admit that I come from Perl - where the rules for where you put your packages, and how the directories should look like is a bit looser, so my point of view my be not so pythonic.
Also, this is a different version of a question I've asked a couple of times here and here and I keep getting answers (that I am grateful for) that say I should not, but never "it's impossible" or "here's how you do it, though you should not". If something is possible, but not advisable, I would rather be able to find out why by trying and seeing and if it is impossible by design, I'd like to know (and stop banging my head against a wall).
Apologies in advance for the long intro - here's the question, finally.
I have this nice module - svgpathtools - that contains methods, classes and sits in an install-specific directory (inside the Python install that comes with Blender), directory that looks like this:
/home/simone/blender/blender-2.92.0-linux64/2.92/python/lib/python3.7/site-packages/svgpathtools
├── __init__.py
├── __pycache__
│   ├── __init__.cpython-37.pyc
│   ├── bezier.cpython-37.pyc
│   ├── document.cpython-37.pyc
│   ├── misctools.cpython-37.pyc
│   ├── parser.cpython-37.pyc
│   ├── path.cpython-37.pyc
│   ├── paths2svg.cpython-37.pyc
│   ├── polytools.cpython-37.pyc
│   ├── smoothing.cpython-37.pyc
│   ├── svg_io_sax.cpython-37.pyc
│   └── svg_to_paths.cpython-37.pyc
├── bezier.py
├── document.py
├── misctools.py
├── parser.py
├── path.py
├── paths2svg.py
├── polytools.py
├── smoothing.py
├── svg_io_sax.py
└── svg_to_paths.py
1 directory, 22 files
I want to add functionality to it without messing with that directory and without subclassing. The first for obvious reasons, the second because for now I am only adding a couple of methods max, so it's not worth maintaining a whole subclass (for now at least). So my idea is to have a directory that looks like this:
/home/simone/blender_learning/mesh-20210920
└── svgpathtools
├── __init__.py
└── distance.py
Now: if I keep svgpathtools/__init__.py, python (3.8.10) won't find the original svgpathtools in the Blender install subdirectory, so all its functionality is gone
and if I remove svgpathtools/__init__.py, python won't find svgpathtools/distance.py so import svgpathtools.distance throws an error.
Which one is it then:
it's impossible: because of the way python looks for packages, you can't have sub-packages in a different place than the main package
I am doing it wrong in some way (please tell me why)
Pretty please: don't just tell me I should not do it without having told me which one it is. I know that doable <> advisable (playing with explosives? petting hungry leopards? eating fugu?)
When you import a module import <module> python looks through the paths for the first match
import sys
print(sys.path)
So having a module with the same name ends up shadowing it, rather than what you want which is keep looking and merge them. So you'll have to name your extensions package something else to not conflict with it.
You can probably get the behaviour you want by rolling your own import method with importlib and a wrapper around the module to fallback to the other but it seems hardly worth it.
Use e.g. sys.path.append("..") to add the directory above the script to the python modules path.
It's not impossible but yes: it's explosive. And that's because of the way python stores its references to the imported modules. If you do:
import sys
print(sys.modules)
you'll see that those references are stored in a dictionnary, and that the keys are the modules names.
Although you could find a messy way to overthrow it, the simplest way would probably using a new name for your own lib, something like svgpathtools_ or svgpathtools_extended for example.

Trying to run python code from repo without instructions

I am trying to run a code from this repo without success. There are no instructions on how to run it. I suspect I should run FactcheckingRANLP/Factchecking_clean/classification/lstm_train.py and then run .../lstm_test.py.
The problem is that this code uses import statements as a module, referencing to folders and files that are in different directories, for example, in lstm_train.py:
File "lstm_train.py", line 3, in <module>
from classification.lstm_utils import *
ModuleNotFoundError: No module named 'classification'
This is the tree structure of the classification folder:
.
├── classification
│   ├── __init__.py
│   ├── __init__.pyc
│   ├── lstm_repres.py
│   ├── lstm_test.py
│   ├── lstm_train.py
│   ├── lstm_train.pyc
│   ├── lstm_utils.py
│   ├── lstm_utils.pyc
│   ├── __pycache__
│   │   ├── __init__.cpython-36.pyc
│   │   ├── lstm_train.cpython-36.pyc
│   │   └── lstm_utils.cpython-36.pyc
│   └── svm_run.py
I would like to know how can I make python run lsmt_train/test.py files in such a manner that the import statements contained within them are compiled correctly. I prefer not to modify the code as this could possibly generate a lot of errors..
You could add the path pointing to the classification folder to your python path variable.
I suggest using the sys package:
import sys
sys.path.append('<<<PathToRepo>>>/FactcheckingRANLP/Factchecking_clean')
With the repo classification directory added to your python path, the import statements should work.
EDIT:
Correction; in the initial post I suggested adding the path to .../classification to your path variable, instead the parent folder .../Factchecking_clean is required as the file imports the module 'classification'.
Also, in Lucas Azevedo's answer, the parent directory path is added in the repository lstm_train file itself. While this definitely works, I still think it should be possible without editing the original repository.
I took a look at the repo in question and files like lstm_train.py are scripts which should be executed with the python working directory set as '<<<PathToRepo>>>/FactcheckingRANLP/Factchecking_clean'.
There are a few ways to do so:
You could open the project in a python IDE and configure your execution to use the directory .../Factchecking_clean as the working directory. In pycharm for example this could be done by importing the repo directory .../Factchecking_clean as a project. The following image shows how to set a working directory for execution in pycharm:
I think the repository was developed with this execution configuration set up.
Another possibility is to execute the python script from within another python file. This seems to be rather inconvenient to me, regardless you could do so by creating a separate python file with:
import sys
import os
sys.path.append('<<<PathToRepo>>>/FactcheckingRANLP/Factchecking_clean')
os.chdir('<<<PathToRepo>>>/FactcheckingRANLP/Factchecking_clean')
exec(open('./classification/lstm_train.py').read())
This adds the Factchecking_clean directory to the python path (using sys.path.append()) to be able to import stuff like classification.utils. The working directory is set by os.chdir() and finally exec(open('<<<filepath>>>')).read() executes the lstm_train file with the correct working directory and path variable set up.
Executing the new python file with the code above works for me (without editing the original repository).
However, as scripts like lstm_train.py actually are used to execute specific parts of the code provided in the rest of the repository modules, I think editing these files for experimental purposes is fine. In general, when working with repositories like this I recommend using an IDE (like pycharm) with a correctly set up configuration (method 1).
Ended up having to modify the repository's code, following the suggestion of Jeff.
import sys,os
parent_dir_path = os.path.abspath(__file__+"/../..")
sys.path.append(parent_dir_path)
adding the path until classification doesnt work, as the import is done mentioning the classification folder. The parent directory is the one that has to be added.
__file__ gives the current file name and os.path.abspath resolves the path navigation done with /../..

unit test structure and import

I want to know how to properly import file in my test file without using __init__.py in test root folder. Last sentence of this artice states, that test directory should not have init file.
I don't know much about python so I would like to know:
1) Why not
2) How to import file tested.py to the test_main.py in order to test its functionality without using init file as a script that insert paths to PYTHONPATHS?
My project has a following structure
.
├── __init__.py
├── my
│   ├── __init__.py
│   ├── main.py
│   └── my_sub
│   ├── __init__.py
│   └── tested.py
└── test
└── test_main.py
Files contains following code
#/Python_test/__init__.py
import my.main
#/Python_test/my/__init__.py
#empty
#/Python_test/my/main.py
from .my_sub.tested import inc
print(inc(5))
#/Python_test/my/my_sub/__init__.py
#empty
#/Python_test/my/my_sub/tested.py
def inc(x):
return x+1
#/Python_test/test/test_main.py
from Python_pytest.my.my_sub.tested import func
def test_answer():
assert func(3) == 3
When I run the code from command line python __init__.py it prints 6, which is correct.
I would like to test the function inc() in tested.py file.
1. I installed pytest package as a testing framework,
2. created test file similar to the one from tutorial here called test_main.py.
3. Added __init__.py with a code that finds path of the root directory and adds it to sys.path
It worked well but then I read that this shouldn't be done this way. But how should it be done? I have hard time reading some unit tested code from some github repositories that are tested and don't use the init file. (one of them is boto3) I can't find any clue that suggest how to properly use it.
I also tried to use relative imports this way
from ..my.main import func
but it throws ValueError: attempted relative import beyond top-level package. Which is ok. But I tried it anyway.
Now I don't know how to do that. Tutorials concerning importing usually states that we should add paths of imported modules to sys.path (if they are not present already) but how should I do that when there shouldn't be the init file which can hold the functionality?

Python can't import my package

I have the following directory structure:
myapp
├── a
│   ├── amodule.py
│   └── __init__.py
├── b
│   ├── bmodule.py
│   ├── __init__.py
└── __init__.py
In a/amodule.py
I have this snippet which calls a simple function in b/bmodule.py
from myapp.b import bmodule
b.myfunc()
But when i run python a/amodule.py I get this error:
File "a/amodule.py", line 1, in <module>
from myapp.b import bmodule
ImportError: No module named 'myapp'
What am I doing wrong?
you need to put your project root onto your python path
you can set the PYTHONPATH environmental variable
or you can alter sys.path before importing
or you can use an IDE like pycharm that will do this kind of thing for you
(although it will probably be from b import blah)
there is likely other ways to resolve this issue as well
watch out for circular imports ...
(in python 3 you can also do relative imports... although I am not a big fan of this feature)
from ..b import blah
the best way to allow
from myapp.b import whatever
would be to edit your .bashrc file to always add your parent path to the PYTHONPATH
export PYTHONPATH=$PYTHONPATH;/home/lee/Code
now every time you log into the system python will treat your Code folder as a default place to look for import modules, regardless of where the file is executed from

Import issues in Python

I am having an import error problem in a test script. It looks like its something to do with the directory structure.
I have the following folder structure:
A
├── F1
│   ├── __init__.py
│   └── Src
│   └── F2
│   └── __init__.py
└── tests1
└── tests1
└── test_script.py
A/F1/Src/F2
F1 has "__init__py" in its level
F2 has "__init__.py" in its level
In the same level as F1, there is another folder "tests1"
tests1/tests1/test_script.py
in test_script.py, I have a line which says
from F1.src.F2 import C
With the above, I get an error saying, no module named "F1.src.F2"
Does someone know what is going on here?
from F1.src.F2 import C is an absolute import. In order for it to work, "F1" has to be located somewhere on your Python path (sys.path). Generally this also includes the current directory if you ran Python on the command line.
So if the A directory is not one of the directories on your Python path and is not your current working directory, there is no reason the import would work.
Module names are case sensitive. You have Src in one place and src in another, but I'm not sure that reflects your actual directory structure or just what you typed here.
Using a relative import will not work if you are running test_script.py as a script (Which is what it sounds like.) So, what you really want to do is make sure that either you run the script from the A directory, or go whole hog, make your project into a legit package with a setup.py and use a test runner such as tox.
I just had to create a shared library with the "egg" file.
As simple as that but it occurred to me late!

Categories