cupy加速numpy

1.安装cupy

查看cuda版本

cuda=11.1

pip3 install cupy-cuda111 -i https://mirrors.aliyun.com/pypi/simple/

2.将在gpu上的<torch.Tensor>数据类型转化成numpy，并放于cpu上

f_cls_sthree=f_cls_sthree.cpu().detach().numpy()
f_cls_tthree = f_cls_tthree.cpu().detach().numpy()

#转化成cupy.ndarray
f_cls_sthree = cp.asarray(f_cls_sthree)
f_cls_tthree = cp.asarray(f_cls_tthree)

3.将np改成cp

def compute_kernel_bias(vecs):
    """计算kernel和bias
    vecs.shape = [num_samples, embedding_size]，
    最后的变换：y = (x + bias).dot(kernel)
    """
    mu = vecs.mean(axis=0, keepdims=True)
    cov = cp.cov(vecs.T)
    u, s, vh = cp.linalg.svd(cov)
    W = np.dot(u, cp.diag(1 / cp.sqrt(s)))
    return W, -mu

def transform_and_normalize(vecs, kernel=None, bias=None):
    """应用变换，然后标准化
    """
    if not (kernel is None or bias is None):
        vecs = (vecs + bias).dot(kernel)
    return vecs / (vecs**2).sum(axis=1, keepdims=True)**0.5

4.将数据重新放入gpu

s_data = torch.tensor(s_data).to(torch.float32).to("cuda")
t_data = torch.tensor(t_data).to(torch.float32).to("cuda")