LWC 68: 766. Toeplitz Matrix

本文介绍了一种判断矩阵是否为Toeplitz矩阵的方法,并提供了Java和Python两种语言的实现代码。Toeplitz矩阵的特点是从左上到右下的每条对角线上的元素都相同。

摘要生成于 C知道 ,由 DeepSeek-R1 满血版支持, 前往体验 >

LWC 68: 766. Toeplitz Matrix

传送门:766. Toeplitz Matrix

Problem:

A matrix is Toeplitz if every diagonal from top-left to bottom-right has the same element.

Now given an M x N matrix, return True if and only if the matrix is Toeplitz.

Example 1:

Input: matrix = [[1,2,3,4],[5,1,2,3],[9,5,1,2]]
Output: True
Explanation:
1234
5123
9512

In the above grid, the diagonals are “[9]”, “[5, 5]”, “[1, 1, 1]”, “[2, 2, 2]”, “[3, 3]”, “[4]”, and in each diagonal all elements are the same, so the answer is True.

Example 2:

Input: matrix = [[1,2],[2,2]]
Output: False
Explanation:
The diagonal “[1, 2]” has different elements.

Note:

  • matrix will be a 2D array of integers.
  • matrix will have a number of rows and columns in range [1, 20].
  • matrix[i][j] will be integers in range [0, 99].

思路:
对角线遍历,注意对角线的性质:当前元素为matrix[i][j], 下一元素为matrix[i+1][j+1]。

代码如下:

    public boolean isToeplitzMatrix(int[][] matrix) {
        int n = matrix.length;
        int m = matrix[0].length;

        for (int j = 0; j < m; ++j) {
            int i = 0;
            while (i + 1 < n && j + 1 < m) {
                if (matrix[i + 1][j + 1] != matrix[i][j]) return false;
                i ++;
                j ++;
            }
        }

        for (int i = 0; i < n; ++i) {
            int j = 0;
            while (i + 1 < n && j + 1 < m) {
                if (matrix[i + 1][j + 1] != matrix[i][j]) return false;
                i ++;
                j ++;
            }
        }
        return true;
    }

当然你可以直接遍历,只要判断每个元素所在对角线的下一个元素是否与当前元素相等即可。

Java版本:

    public boolean isToeplitzMatrix(int[][] matrix) {
        int n = matrix.length;
        int m = matrix[0].length;
        for (int i = 0; i < n - 1; ++i) {
            for (int j = 0; j < m - 1; ++j) {
                if (matrix[i][j] != matrix[i + 1][j + 1]) return false;
            }
        }
        return true;
    }

Python版本:

    def isToeplitzMatrix(self, matrix):
        """
        :type matrix: List[List[int]]
        :rtype: bool
        """
        n = len(matrix)
        m = len(matrix[0])

        for i in range(0, n - 1):
            for j in range(0, m - 1):
                if matrix[i][j] != matrix[i + 1][j + 1]: return False
        return True
class UniformAffineQuantizer(nn.Module): def __init__( self, n_bits: int = 8, symmetric: bool = False, per_channel_axes=[], metric="minmax", dynamic=False, dynamic_method="per_cluster", group_size=None, shape=None, lwc=False, disable_zero_point=False, ): """ support cluster quantize dynamic_method support per_token and per_cluster """ super().__init__() self.symmetric = symmetric self.disable_zero_point = disable_zero_point assert 2 <= n_bits <= 16, "bitwidth not supported" self.n_bits = n_bits if self.disable_zero_point: self.qmin = -(2 ** (n_bits - 1)) self.qmax = 2 ** (n_bits - 1) - 1 else: self.qmin = 0 self.qmax = 2 ** (n_bits) - 1 self.per_channel_axes = per_channel_axes self.metric = metric self.cluster_counts = None self.cluster_dim = None self.scale = None self.zero_point = None self.round_zero_point = None self.cached_xmin = None self.cached_xmax = None self.dynamic = dynamic self.dynamic_method = dynamic_method self.deficiency = 0 self.lwc = lwc init_value = 4. # inti value of learnable weight clipping if lwc: if group_size: dim1 = int(shape[0]*math.ceil(shape[1]/group_size)) self.deficiency = shape[-1]%group_size if self.deficiency > 0: self.deficiency = group_size - self.deficiency assert self.symmetric # support for mlc-llm symmetric quantization else: dim1 = shape[0] self.upbound_factor = nn.Parameter(torch.ones((dim1,1))*init_value) self.lowbound_factor = nn.Parameter(torch.ones((dim1,1))*init_value) self.sigmoid = nn.Sigmoid() self.enable = True self.group_size = group_size def change_n_bits(self, n_bits): self.n_bits = n_bits if self.disable_zero_point: self.qmin = -(2 ** (n_bits - 1)) self.qmax = 2 ** (n_bits - 1) - 1 else: self.qmin = 0 self.qmax = 2 ** (n_bits) - 1 def fake_quant(self, x, scale, round_zero_point): if self.deficiency > 0: pad_zeros = torch.zeros((x.shape[0],self.deficiency),dtype=x.dtype,device=x.device) x = torch.cat((x,pad_zeros),dim=1) if self.group_size: assert len(x.shape)==2, "only support linear layer now" dim1, dim2 = x.shape x = x.reshape(-1, self.group_size) x_int = round_ste(x / scale) if round_zero_point is not None: x_int = x_int.add(round_zero_point) x_int = x_int.clamp(self.qmin, self.qmax) x_dequant = x_int if round_zero_point is not None: x_dequant = x_dequant.sub(round_zero_point) x_dequant = x_dequant.mul(scale) if self.group_size: x_dequant = x_dequant.reshape(dim1, dim2) if self.deficiency > 0: x_dequant = x_dequant[:,:-self.deficiency] return x_dequant def forward(self, x: torch.Tensor): if self.n_bits >= 16 or not self.enable: return x if self.metric == "fix0to1": return x.mul_(2**self.n_bits-1).round_().div_(2**self.n_bits-1) if self.dynamic_method == "per_token" or self.dynamic_method == "per_channel": self.per_token_dynamic_calibration(x) else: raise NotImplementedError() x_dequant = self.fake_quant(x, self.scale, self.round_zero_point) return x_dequant def per_token_dynamic_calibration(self, x): if self.group_size: if self.deficiency == 0: x = x.reshape(-1,self.group_size) else: pad_zeros = torch.zeros((x.shape[0],self.deficiency),dtype=x.dtype,device=x.device) x = torch.cat((x,pad_zeros),dim=1) x = x.reshape(-1,self.group_size) reduce_shape = [-1] xmin = x.amin(reduce_shape, keepdim=True) xmax = x.amax(reduce_shape, keepdim=True) if self.lwc: xmax = self.sigmoid(self.upbound_factor)*xmax xmin = self.sigmoid(self.lowbound_factor)*xmin if self.symmetric: abs_max = torch.max(xmax.abs(),xmin.abs()) scale = abs_max / (2**(self.n_bits-1)-1) self.scale = scale.clamp(min=CLIPMIN, max=1e4) zero_point = (2**(self.n_bits-1)-1)*torch.ones_like(self.scale) else: range = xmax - xmin scale = range / (2**self.n_bits-1) self.scale = scale.clamp(min=CLIPMIN, max=1e4) zero_point = -(xmin) / (self.scale) if self.disable_zero_point: self.round_zero_point = None else: self.round_zero_point = zero_point.clamp(min=-1e4, max=1e4).round() def register_scales_and_zeros(self): self.register_buffer('scales', self.scale) self.register_buffer('zeros', self.round_zero_point) del self.scale del self.round_zero_point
最新发布
07-24
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值