CODEWITHSUNDEEP
pythonimageperformancenumpyfor-loop
Yash Bhutada
Yash Bhutada

Reputation: 37

How do I reduce the run time of this code containing 2 for loops?

In the following piece of code, the 2 for loops take about .05s to complete on average. data is a numpy array. Each i, j cell contains a tuple that holds the RGB values as defined by those functions. I am later building up an image using those RGB values and i, j are x, y pixel coordinates.

Is there any way to reduce the time of this operation? Or is there any other faster way that I build up an image by specifying the RGB values of each pixels as some sort of mathematical functions?

start = time.time()
for i in range (0, 150):
    for j in range(0, 150):
        data[i,j] = [int(math.sin(math.pi*i/300.0)*127.5 + 127.5),
                     int(math.cos(peak)*127.5 + 127.5),
                     int(math.sin(math.pi*j/300.0)*127.5 + 127.5)] 
print ('Time: ', time.time() - start)```

Upvotes: 1

Views: 59

Answers (1)

user3483203
user3483203

Reputation: 51185

You can remove all the loops here using ogrid and instead just use three calculations, one for r, g, and b.

x, y, z = data.shape
i, j = np.ogrid[:x, :y]

data[..., 0] = (np.sin(np.pi*i/300)*127.5 + 127.5).astype(int)
data[..., 1] = (np.cos(peak)*127.5 + 127.5).astype(int)
data[..., 2] = (np.sin(np.pi*j/300)*127.5 + 127.5).astype(int)

Performance

def rgb_vectorized(x, y, peak=1):
    data = np.empty((x, y, 3), dtype=int)
    i, j = np.ogrid[:x, :y]
    data[..., 0] = (np.sin(np.pi*i/300)*127.5 + 127.5).astype(int)
    data[..., 1] = (np.cos(peak)*127.5 + 127.5).astype(int)
    data[..., 2] = (np.sin(np.pi*j/300)*127.5 + 127.5).astype(int)
    return data

def rgb_original(x, y, peak=1):
    data = np.empty((x, y, 3), dtype=int)
    for i in range (x):
        for j in range(y):
            data[i,j] = [int(math.sin(math.pi*i/300.0)*127.5 + 127.5),
                         int(math.cos(peak)*127.5 + 127.5),
                         int(math.sin(math.pi*j/300.0)*127.5 + 127.5)]
    return data

%timeit rgb_vectorized(1000, 1000)
9.85 ms ± 109 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%timeit rgb_original(1000, 1000)
3.4 s ± 27.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Validation

>>> np.array_equal(rgb_vectorized(1000, 1000), rgb_original(1000, 1000))
True

Upvotes: 3

Related Questions