For this (and similar) use case, I would recommend a combination of pip and virtualenv.
You would install pip into your system Python install (i.e. sudo apt-get install python-pip), and then install virtualenv via pip, i.e. pip install virtualenv).
You can then create a specific virtualenv for this project. This represents a sandboxed environment with specific versions of libraries that are specified traditionally through a requirements file (using the -r option), but can also be specified individually through the command line.
You would do this via command like virtualenv venv_test, which will create a virtualenv directory named venv_test in the current directory. You can then run pip from that virtualenv's bin dir to install packages.
For example, to install the flask package in that virutalenv, you would run:
venv_test/bin/pip install flask
You can then either run source venv_test/bin/activate to put the current shell into the virtualenv's, or invoke a script directly from the virtualenv's interpreter, i.e.:
venv_test/bin/python foo.py
Here's link to a virtualenv introduction for some additional details/steps.
setuptools(pypi.python.org/pypi/setuptools). After you install it, just runeasy_install Numpyin your shell script.softenv. Maybe talk to the cluster admin about it?