How do I parallelize this using Python multiprocessing?

Question

I've coded a 1 vs rest classifier in Python that trains 11 different classifiers, one for each class. The code is shown below:

def onevsrest(X_train,y_train,lamb):
    beta=[]
    beta_init=np.zeros(X_train.shape[1])
    for i in range(1,12):
        print(i)
        y=np.copy(y_train)
        y[y != i] = -1
        y[y == i] = 1
        beta_temp,objs = svm(lamb, 0.1, 200, X_train, y)
        beta.append(beta_temp[-1])
    return beta

How would I parallelize the above program using the Python multiprocessing module? From my understanding, multiprocessing can only be used for code with a single argument. How would I scale that towards this function that takes multiple arguments?

Answer 1

You can "partial" your function. Example:

# multiple arguments function
def calc(a, b, c):
    return a + b + c

# prepare a single argument partial function, freezing `b` and `c`
from functools import partial
calc2 = partial(calc, b=3, c=7)

from multiprocessing import Pool
p = Pool(5)
print(p.map(calc2, [1, 2, 3, 4, 5, 6]))

Answer 2

You could also use tuple , like:

from multiprocessing import Pool

def f(data):
    x, y = data
    return x*y

if __name__ == '__main__':
    with Pool(5) as p:
        X = (1,2,3)
        Y = (1,4,9)
        print(p.map( f, list(zip(X,Y)) )) # returns [1, 8, 27]