1

我的应用程序需要存储和访问一个非常大的矩阵。我虽然会为此使用 dok_matrix ,但我似乎无法对文件进行内存映射。我收到错误:数组无法进行内存映射:dtype 中的 Python 对象。

这是代码:

def __init__(self, ratings_file):

    try:
        self.ratings = np.load(ratings_file, mmap_mode='r+')
        self.n_users = self.ratings.shape[0]
        self.n_items = self.ratings.shape[1]
    except IOError:
        self.ratings = None
        self.n_users = 0
        self.n_items = 0

def add_ratings(self, ratings):
    # update the number of users and items
    self.n_users = max(self.n_users, max([r[0] for r in ratings]) + 1)
    self.n_items = max(self.n_items, max([r[1] for r in ratings]) + 1)

    # reshape (or create) the matrix
    if not self.ratings:
        self.ratings = dok_matrix((self.n_users, self.n_items), dtype=np.dtype(np.float32))
    else:
        self.ratings.resize((self.n_users, self.n_items))

    print 'Num. users: ', self.n_users
    print 'Num. items: ', self.n_items

def update_model(self):
    raise NotImplementedError()

def save(self):
    np.save(self._ratings_file, self.ratings)

有什么帮助吗?谢谢

4

1 回答 1

0

在这一行中 self.ratings = dok_matrix((self.n_users, self.n_items), dtype=np.dtype(np.float32)) 为什么type等于这个复合事物而不是 just np.float32?你不是只是想说你想要一个浮点矩阵吗?

于 2012-12-10T14:47:56.173 回答