1. NeSI Support
  2. Scientific Computing
  3. Supported Applications

Python

Last updated  

Licences

All versions of Python available on NeSI platforms are owned and licensed by the Python Software Foundation. Each version is released under a specific open-source licence. The licences are available on the Python documentation server.

System Python vs Environment Modules

Our operating systems include Python but not an up to date version, so we strongly recommend that you load one of our Python environment modules instead.  They include optimised builds of the most popular Python packages for computational work such as numpy, scipy, matplotlib, and many more.

NeSI Customisations

Our most recent Python environment modules have:

  • multiprocessing.cpu_count() patched to return only the number of CPUs available to the process, which in a Slurm job can be fewer than the number of CPUs on the node.

  • PYTHONUSERBASE set to a path which includes the toolchain, so that incompatible builds of the same version of Python don't attempt to share user-installed libraries.

Example scripts

#!/bin/bash -e

#SBATCH --job-name    MyPythonJob
#SBATCH --time        01:00:00
#SBATCH --mem         512MB

module load Python/3.7.3-gimkl-2018b

python MyPythonScript.py

MPI Example

  #!/bin/bash -e
  #SBATCH --job-name=PythonMPI
  #SBATCH --ntasks=2          # Number of MPI tasks
  #SBATCH --time=00:30:00
  #SBATCH --mem-per-cpu=512MB # Memory per logical CPU

  module load Python
  srun python PythonMPI.py   # Executes ntasks copies of the script
import numpy as np
from mpi4py import MPI

comm = MPI.COMM_WORLD
size = comm.Get_size() # Total number of MPI tasks
rank = comm.Get_rank() # Rank of this MPI task

# Calculate the data (numbers 0-9) on the MPI ranks
rank_data = np.arange(rank, 10, size)

# perform some operation on the ranks data
rank_data += 1

# gather the data back to rank 0
data_gather = comm.gather(rank_data, root = 0)

# on rank 0 sum the gathered data and print both the sum of, 
# and the unsummed data
if rank == 0:
    print('Gathered data:', data_gather)
    print('Sum:', sum(data_gather))

The above Python script will create a list of numbers (0-9) split between the MPI tasks (ranks). Each task will then add one to the numbers it has, those numbers will then be gathered back to task 0, where the numbers will be summed and both the sum of, and the unsummed data is printed.

Multiprocessing Example

  #!/bin/bash -e
  #SBATCH --job-name=PytonMultiprocessing
  #SBATCH --cpus-per-task=2   # Number of logical CPUs
  #SBATCH --time=00:10:00
  #SBATCH --mem-per-cpu=512MB # Memory per logical CPU

  module load Python
  python PythonMultiprocessing.py
import multiprocessing

def calc_square(numbers, result1):
    for idx, n in enumerate(numbers):
        result1[idx] = n*n

def calc_cube(numbers, result2):
    for idx, n in enumerate(numbers):
        result2[idx] = n*n*n

if __name__ == "__main__":
    numbers = [2,3,4]
    # Sets up the shared memory variables, allowing the variables to be
    # accessed globally across processes
    result1 = multiprocessing.Array('i',3)
    result2 = multiprocessing.Array('i',3)
    # set up the processes
    p1 = multiprocessing.Process(target=calc_square, args=(numbers,result1,))
    p2 = multiprocessing.Process(target=calc_cube, args=(numbers,result2,))

    # start the processes
    p1.start()
    p2.start()

    # end the processes
    p1.join()
    p2.join()

    print(result1[:])
    print(result2[:])

The above Python script will calculated the square and cube of an array of numbers using multiprocessing and print the results from outside of those processes, safely circumventing Python's global interpreter lock.

For more in depth examples of and descriptions of Multiprocessing in Python you may find this Multithreading/Multiprocessing Youtube tutorial series helpful

Job Arrays

Job arrays can be handled using the Slurm environment variable SLURM_ARRAY_TASK_ID as array index. This index can be called directly from within the script or using a command line argument. In the following both options are presented:

The job scripts calling both examples:

#!/bin/bash -e
#SBATCH -J test
#SBATCH --time=00:01:00
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --array=1-2 # Array jobs

module load Anaconda3

echo "SLURM_ARRAY_TASK_ID $SLURM_ARRAY_TASK_ID of $SLURM_ARRAY_TASK_COUNT"

#env variable in python
python hello_world.py

#as command line argument
python hello_world_args.py -ID $SLURM_ARRAY_TASK_ID

the version getting the env variable in the python script "hello_world.py"

#!/usr/bin/env python3

import os
my_id = os.environ['SLURM_ARRAY_TASK_ID']
print("hello world with ID {}".format(my_id))

the version getting the env variable as argument in the python script "hello_world_args.py"

#!/usr/bin/env python3
"""
Module for handling input arguments
"""

import argparse

# get tests from file
class LoadFromFile(argparse.Action):
    """
    class for reading arguments from file
    """
    def __call__(self, parser, namespace, values, option_string=None):
        with values as F:
            vals = F.read().split()
        setattr(namespace, self.dest, vals)

def get_args():
    """
    Definition of the input arguments
    """
    parser = argparse.ArgumentParser(description='Hello World')
    parser.add_argument('-ID', type=int, action='store', dest='my_id',
                        help='Slurm ID')
    return parser.parse_args()


ARGS = get_args()
print("hello world from ID {}".format(ARGS.my_id))

Python Packages

Programmers around the world have written and released many packages for Python, which are not included with the core Python distribution and must be installed separately. Each Python environment module comes with its own particular suite of packages, and the system Python has its own installed packages.

The provided packages can be listed using

module load Python/3.10.5-gimkl-2022a
python -c "help('modules')"

Installing packages in your $HOME

This is the simplest way to install additional packages, but you might fill your $HOME quota and cannot share installations with collaborators.

module load Python/3.10.5-gimkl-2022a
pip install --user prodXY

If you are working on multiple projects, this method will cause issues as your projects may require different versions of packages which are not compatible.

We strongly recommend using separate Python virtual environments to isolate dependencies between projects, avoid filling your home space and being able to share installation with collaborators

Installing packages in a Python virtual environment

A Python virtual environment is lightweight system to create an environment which contains specific packages for a project, without interfering with the global Python installation. Each virtual environment is a different directory.

To create a Python virtual environment, use the venv module as follows

module load Python/3.10.5-gimkl-2022a
python3 -m venv /nesi/project/PROJECT_ID/my_venv

where PROJECT_ID is your NeSI project ID.

Note that you need to activate the virtual environment before using it (to run a script or install packages)

source /nesi/project/PROJECT_ID/my_venv/bin/activate

Then you can install any pip-installable package in the virtual environment using

pip install prodXY

Then a Slurm job submission script running your Python script would look like

#!/bin/bash -e
#SBATCH --job-name    MyPythonJob
#SBATCH --time        01:00:00
#SBATCH --mem         512MB

module purge
module load Python/3.10.5-gimkl-2022a
source /nesi/project/PROJECT_ID/my_venv/bin/activate
python MyPythonScript.py

Python virtual environment isolation

By default, Python virtual environments are fully isolated from the system installation. It means that you will not be able to access packages already prepared by NeSI in the corresponding Python environment module.

To avoid this, use the option --system-site-packages when creating the virtual environment

module load Python/3.10.5-gimkl-2022a
python3 -m venv --system-site-packages /nesi/project/PROJECT_ID/my_venv

A downside of this is that now your virtual environment also finds packages from your $HOME folder. To avoid this behavirour, make sure to use export PYTHONNOUSERSITE=1 before calling pip or running a Python script. For example, in a Slurm job submission script

#!/bin/bash -e
#SBATCH --job-name    MyPythonJob
#SBATCH --time        01:00:00
#SBATCH --mem         512MB

module purge
module load Python/3.10.5-gimkl-2022a
source /nesi/project/PROJECT_ID/my_venv/bin/activate
export PYTHONNOUSERSITE=1
python MyPythonScript.py

Further notes

iPython

iPython (interactive Python) is an enhanced tool for accessing a Python command line. It is available in many NeSI Python modules.

Starting iPython

To open an iPython console, simply run the ipython command:

[jblo123@build-wm ~]$ module load Python/3.6.3-gimkl-2017a
[jblo123@build-wm ~]$ ipython

Listing available functions

You can use iPython to list the functions available that start with a given string. Please note that if the string denotes a module (i.e., it has a full stop somewhere in it), that module (or the function you want from it) must first be imported, using either an "import X" statement or a "from X import Y" statement.

import os
os.<TAB>   # List all functions in the os module
os.O_<TAB> # List functions starting with "O_" from the os module
len<TAB>   # List functions starting with "len"

Here,

o<TAB>

or even

os<TAB>

and expect to see the methods and values provided by the os module - you have to put the full stop after the "os" if you want to do that.

Getting information about an object

In iPython, you can query any object by typing the object name followed by a question mark (?), then hitting Enter. For instance:

In [1]: x = 5
In [2]: x?
Type:        int
String form: 5
Docstring:
int(x=0) -> int or long
int(x, base=10) -> int or long

Convert a number or string to an integer, or return 0 if no arguments
are given.  If x is floating point, the conversion truncates towards zero.
If x is outside the integer range, the function returns a long instead.

If x is not a number or if base is given, then x must be a string or
Unicode object representing an integer literal in the given base.  The
literal can be preceded by '+' or '-' and be surrounded by whitespace.
The base defaults to 10.  Valid bases are 0 and 2-36.  Base 0 means to
interpret the base from the string as an integer literal.
>>> int('0b100', base=0)
4

You can also do this on functions (len?), methods (os.mkdir?) and modules (os.path?). If you try to do it on something that isn't defined yet, Python will tell you that the object in question couldn't be found.

Quitting iPython

Just enter the quit command at the iPython prompt.

Labels: mahuika app
Was this article helpful?
0 out of 0 found this helpful