CODEWITHSUNDEEP
pythonfunctionclassmethods
Yudhishthira Dattachaudhuri
Yudhishthira Dattachaudhuri

Reputation: 27

Python class method runs much slower than identical function

I have the following class:

class MyKMeans:
    def __init__(self, max_iter = 300):
        self.max_iter = max_iter
        # Directly access 
        self.centroids = None
        self.clusters = None

    def fit(self, X, k):
        """
        
        """

        # each point is assigned to a cluster
        clusters = np.zeros(X.shape[0])

        # select k random centroids
        random_idxs = np.random.choice(len(X), size=k, replace=False)
        centroids = X[random_idxs, :]

        # iterate until no change occurs in centroids
        while True: 
            # for each point
            for i, point in enumerate(X):
                min_d = float('inf')

                # find the closest centroid to the point
                for idx, centroid in enumerate(centroids):
                    d = euclidean_dist(centroid, point)
                    if d < min_d:
                        min_d = d
                        clusters[i] = idx

                # update the new centroids by averaging the points in each cluster
                new_centroids = pd.DataFrame(X).groupby(by=clusters).mean().values
            
            # if the centroids didn't change, then stop
            if np.count_nonzero(centroids-new_centroids) == 0:
                break
            # otherwise, update the centroids
            else:
                centroids = new_centroids

        self.centroids = centroids
        self.clusters = clusters

and run it using

k = 4
kmeans = MyKMeans()
kmeans.fit(X, k)
centroids, clusters = kmeans.centroids, kmeans.clusters

However, this takes usually 5 seconds to complete running. On the other hand, if I move the method to a new function,

def fit(X, k):
    """
    
    """

    # each point is assigned to a cluster
    clusters = np.zeros(X.shape[0])

    # select k random centroids
    random_idxs = np.random.choice(len(X), size=k, replace=False)
    centroids = X[random_idxs, :]

    # iterate until no change occurs in centroids
    while True: 
        # for each point
        for i, point in enumerate(X):
            min_d = float('inf')

            # find the closest centroid to the point
            for idx, centroid in enumerate(centroids):
                d = euclidean_dist(centroid, point)
                if d < min_d:
                    min_d = d
                    clusters[i] = idx

            # update the new centroids by averaging the points in each cluster
            new_centroids = pd.DataFrame(X).groupby(by=clusters).mean().values
        
        # if the centroids didn't change, then stop
        if np.count_nonzero(centroids-new_centroids) == 0:
            break
        # otherwise, update the centroids
        else:
            centroids = new_centroids

        return centroids, clusters

and get the same variables by calling centroids, clusters = fit(X, k), the runtime is around 0.5-1 second which is a big difference.

Is there a reason why simply having a class method instead of a function causes such a big difference in runtime, and is there any way to improve the runtime while still being able to use the class?

Upvotes: 1

Views: 346

Answers (1)

Wolph
Wolph

Reputation: 80031

The return statement in your non-class version is inside of the while loop so it exits the loop early.

Upvotes: 2

Related Questions