The numpy git repo includes a useful allocation_tracking package in the tools directory, but numpy doesn't install it by default. This repo includes a conda recipe to install that package, along with some extra utilities and decorators.

A Tiny Note: The allocation_tracking module's setup.py would normally install alloc_hook into the global namespace, but this package installs everything into the namespace numpy_allocation_tracking.

conda install -c ilastik numpy-allocation-tracking

To build this recipe yourself for say, numpy 1.9:

conda build --numpy=1.9 conda-recipe

Note: If you make any changes to this repo, you must commit them before building the recipe. Uncommitted changes are not included in the build.

from numpy_allocation_tracking.decorators import assert_mem_usage_factor

# If this function uses more RAM than 2x the size
# of input_array, raise an error.
# (Useful for testing purposes, not in production code.)
@assert_mem_usage_factor(2.0)
def dumb_function(input_array, x, y, z):
    return x * (input_array + y + z)

result = dumb_function( np.ones((100,100)), 10, 20, 30 )