Collaborative Filtering for row items in a given N-by-M matrix

collaborative_filtering.py

import numpy as np
    
class CF:
    """Collaborative filtering for row items in a given N-by-M matrix
    """
    
    def __init__(self, mat):
    	self.mat = mat
    	self.N, self.M = mat.shape
    	
    def similarity(self, v1, v2, func='cosine'):
    	"""Compute similarity between v1 and v2
    	v1 and v2 are vectors (1D numpy array)
    	"""
    	
    	if np.array_equal(v1, v2): return 1.
    	
    	if func == 'cosine':
            v1_norm = np.sqrt(sum(v1 ** 2))
            v2_norm = np.sqrt(sum(v2 ** 2))
            if v1_norm == 0. or v2_norm == 0.: return 0.
            return float(sum(v1 * v2)) / (v1_norm * v2_norm)
    	elif func == 'euclidean':
            return 1. / (1. + np.sqrt(sum((v1 - v2) ** 2)))
    	
    def run(self):
    	self.sims_mat = np.zeros((self.N, self.N))
    	for i in xrange(self.N):
    	    for j in xrange(i, self.N):
                sim = self.similarity(self.mat[i, :], self.mat[j, :])
                self.sims_mat[i, j] = self.sims_mat[j, i] = sim
    		
    def top_n(self, n):
    	"""Return Top-N similar items' indices for each row
    	"""
    	if not hasattr(self, 'sims_mat'): self.run()
    	tops_idx = self.sims_mat[0, :].argsort()[::-1][:n]
    	tops_sim = np.sort(self.sims_mat[0, :])[::-1][:n]
    	for i in xrange(1, self.N):
    	    tops_idx = np.vstack((tops_idx, self.sims_mat[i, :].argsort()[::-1][:n]))
    	    tops_sim = np.vstack((tops_sim, np.sort(self.sims_mat[i, :])[::-1][:n]))
    	return tops_idx, tops_sim