def ret80000(): path = '/media/haboric/Ubuntu Data/yizhou/arctic-captions/json/real' sp = [] #for i in range(1,9): for i in range(1,8): with open(path+'/train'+str(i*10000)+'.nd', 'rb') as f: sp.append(csr_matrix((lo(f),lo(f),lo(f)))) #with open(path+'/train82783.nd', 'rb') as f: # sp.append(csr_matrix((lo(f),lo(f),lo(f)))) print 'before reduce' sv = reduce(conc, sp) del sp return sv
def ret80000(): path = '/media/haboric/Ubuntu Data/yizhou/arctic-captions/json/real' sp = [] #for i in range(1,9): for i in range(1, 8): with open(path + '/train' + str(i * 10000) + '.nd', 'rb') as f: sp.append(csr_matrix((lo(f), lo(f), lo(f)))) #with open(path+'/train82783.nd', 'rb') as f: # sp.append(csr_matrix((lo(f),lo(f),lo(f)))) print 'before reduce' sv = reduce(conc, sp) del sp return sv
def conc(matrix1, matrix2): new_data = numpy.concatenate((matrix1.data, matrix2.data)) new_indices = numpy.concatenate((matrix1.indices, matrix2.indices)) new_ind_ptr = matrix2.indptr + len(matrix1.data) new_ind_ptr = new_ind_ptr[1:] new_ind_ptr = numpy.concatenate((matrix1.indptr, new_ind_ptr)) return csr_matrix((new_data, new_indices, new_ind_ptr)) path = 'real' sp = [] for i in range(1, 9): with open(path + '/train' + str(i * 10000) + '.nd', 'rb') as f: sp.append(csr_matrix((lo(f), lo(f), lo(f)))) with open(path + '/train82783.nd', 'rb') as f: sp.append(csr_matrix((lo(f), lo(f), lo(f)))) #print 'before reduce' sv = reduce(conc, sp) del sp #print 'before dump' with open(path + '/coco_align.train.feat.nd', 'wb') as f: pkl.dump(sv, f, protocol=pkl.HIGHEST_PROTOCOL)
def conc(matrix1, matrix2): new_data = numpy.concatenate((matrix1.data, matrix2.data)) new_indices = numpy.concatenate((matrix1.indices, matrix2.indices)) new_ind_ptr = matrix2.indptr + len(matrix1.data) new_ind_ptr = new_ind_ptr[1:] new_ind_ptr = numpy.concatenate((matrix1.indptr, new_ind_ptr)) return csr_matrix((new_data, new_indices, new_ind_ptr)) path = 'real' sp = [] for i in range(1,9): with open(path+'/train'+str(i*10000)+'.nd', 'rb') as f: sp.append(csr_matrix((lo(f),lo(f),lo(f)))) with open(path+'/train82783.nd', 'rb') as f: sp.append(csr_matrix((lo(f),lo(f),lo(f)))) #print 'before reduce' sv = reduce(conc, sp) del sp #print 'before dump' with open(path+'/coco_align.train.feat.nd', 'wb') as f: pkl.dump(sv,f,protocol=pkl.HIGHEST_PROTOCOL)