Process profiler

A simple process profiler. propro can be used in many different ways. Conveniently it can be used on the command line:

$ propro --fmt=png <command>

For more options, call:

$ propro --help

Another option is to call the profiling programmatically:

import propro
x = propro.profile_cmd("ufig --background-type=chunked_map ufig.config.random")

The returned profiling result can then, for instance, be used for custom plotting.

propro offers the option to profile a single Python function using a decorator:

import propro
import numpy as np

@propro.profile(sample_rate=0.1, fmt="txt")
def mem_hungry(size):
    a = []
    for i in range(size):
        a.append(np.random.random())
        
    b = []
    for i in range(size):
        t = []
        for j in range(size):
            t.append(i * a[j])
        b.append(t)

    b = np.array(b)

The profiling output is stored in the folder where the Python code was launched.

Finally, propro can be embedded in your IPython notebooks. Load the extension with:

import propro
%load_ext propro

The profiling can be done on a line level:

%propro -r 0.1 load_pixels(path, PIXEL_COUNT)

or on a cell level:

%%propro -r 0.1
X = np.random.normal(size=(200,200,1000))
P, D, Q = np.linalg.svd(X, full_matrices=False)
X_a = np.dot(np.dot(P, np.diag(D)), Q)
print(np.std(X), np.std(X_a), np.std(X - X_a))

The output will look something like this if rendered into an image:

jakeret/propro

Process profiler