- android - 多次调用 OnPrimaryClipChangedListener
- android - 无法更新 RecyclerView 中的 TextView 字段
- android.database.CursorIndexOutOfBoundsException : Index 0 requested, 光标大小为 0
- android - 使用 AppCompat 时,我们是否需要明确指定其 UI 组件(Spinner、EditText)颜色
我的最终目标是训练一个基于均值和协方差参数化的 4D 多元高斯分布
在哪里,
和,
目前我有以下代码:
import tensorflow as tf
import numpy as np
value = [[1,2.0,3,4,5],[0,2,4,6,8],[80,7,6,5,4]]
value=tf.constant(value)
cov= tf.slice(value,[0,int(value.shape[1])-1],[int(value.shape[0]),1])
mean= tf.slice(value,[0,0],[int(value.shape[0]),int(value.shape[1])-1])
eyes=tf.eye(int(mean.shape[1]),batch_shape=[int(value.shape[0])])
#eyes = tf.multiply(eyes,cov)
normal = tf.contrib.distributions.MultivariateNormalFullCovariance(
loc=mean,
covariance_matrix=eyes)
value = [[1,2.0,3,4,5],[0,2,4,6,8],[80,7,6,5,4]]
是其余代码可能接收的示例。
在上面的例子中
cov = <tf.Tensor 'Slice_2:0' shape=(3, 1) dtype=float32>
eyes = <tf.Tensor 'eye_1/MatrixDiag:0' shape=(3, 4, 4) dtype=float32>
cov = [[5.] [8.] [4.]]`
eyes = [[[1. 0. 0. 0.]
[0. 1. 0. 0.]
[0. 0. 1. 0.]
[0. 0. 0. 1.]]
[[1. 0. 0. 0.]
[0. 1. 0. 0.]
[0. 0. 1. 0.]
[0. 0. 0. 1.]]
[[1. 0. 0. 0.]
[0. 1. 0. 0.]
[0. 0. 1. 0.]
[0. 0. 0. 1.]]]`
我的问题是,如何在给定 cov
和 eyes
的情况下获得 result
?结果如下:
result = [[[5., 0., 0., 0.],
[0., 5., 0., 0.],
[0., 0., 5., 0.],
[0., 0., 0., 5.]],
[[8., 0., 0., 0.],
[0., 8., 0., 0.],
[0., 0., 8., 0.],
[0., 0., 0., 8.]],
[[4., 0., 0., 0.],
[0., 4., 0., 0.],
[0., 0., 4., 0.],
[0., 0., 0., 4.]]]
提前致谢。
最佳答案
Tensorflow 使用与 numpy 相同类型的索引,这可能非常强大。
您可以在这里查看详细信息:https://docs.scipy.org/doc/numpy-1.13.0/user/basics.broadcasting.html请注意,np.newaxis
定义为与 None
相同。
对于您的问题,您可以为数据添加一个额外的维度,以确保数组相乘的方式没有歧义。
import numpy as np
cov = np.array([[5.],[8.],[4.]])
eyes = np.array([[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]],[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]],[[1.0,0,0,0],[0.0,1.0,0.0,0],[0.0,0.0,1.0,0],[0.0,0.0,0.0,1.0]]])
result = cov[:,:,None]*eyes
在这里使用 None
增加了一个额外的维度,使 cov
成为一个 3x1x1 数组,它可以明确地与 3x4x4 数组相乘。您也可以在 tensorflow 中以这种方式使用 None
。
如果每个对应的维度大小相同,或者其中一个维度的大小为 1,则两个数组可以明确相乘。
关于python - 在 Tensorflow 中将一组常量(一维数组)与一组矩阵(三维数组)相乘,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/52087206/
我有一个 Haskell 程序,它用 Metropolis 模拟 Ising 模型 算法。主要操作是一个模板操作,它需要下一个的总和 2D 中的邻居,然后将其与中心元素相乘。那么 元素可能已更新。 在
对于卷积,我想在一个维度上应用圆形填充,在所有其他维度上应用零填充。我该怎么做? 对于卷积,有 28 个 channel ,并且数据在球形箱中描述。半径时间有 20 个箱子,极地时间有 20 个箱子,
假设您有一个数组并想创建另一个数组,该数组的值连续等于第一个数组的 10 个元素的标准差。在 for 循环的帮助下,它可以像下面的代码一样轻松编写。我想要做的是避免使用 for 循环来加快执行时间。有
我有一个长度为size * size的一维数组,表示值的平方字段。 我的目标是将数组旋转到位(previous question)。我目前在获取正确的内圈索引方面遇到问题。我的算法有什么错误? 这是我
stride = 1 和 1 个过滤器的 1D CNN 的输出长度不应该等于输入长度而不需要填充吗? 我认为是这种情况,但创建了一个具有这些规范的 Keras 模型,当输入形状为 (17910,1)
我有一个零散布的一维数组。想创建第二个数组,其中包含最后一个零的位置,如下所示: >>> a = np.array([1, 0, 3, 2, 0, 3, 5, 8, 0, 7, 12]) >>> fo
我试着去理解卡尔曼滤波器是如何工作的,因为多维变量太混乱了,一开始我从一维的例子开始。 我发现有3个不同的来源解释温度计的情况,但所有这些情况实现的方程略有不同,我没有得到重点。 我实现了解决方案2,
我有一个二维数组,我需要将它转换为一个列表(同一对象)。我不想使用 for 或 foreach 循环来执行此操作,它们将获取每个元素并将其添加到列表中。还有其他方法吗? 最佳答案 好吧,你可以让它使用
我正在使用 Keras 构建一个 CNN,将以下 Conv1D 作为我的第一层: cnn.add(Conv1D( filters=512, kernel_size=3, str
在 Python 中使用 TensorFlow,我正在制作一个以一维数组作为输入的神经网络。我想在网络中添加一个卷积层,但似乎无法让它工作。 我的训练数据如下所示: n_samples = 20 le
我有一个简单的一维数组,但我想使用 jQuery 将长列表分成两列。我怎样才能实现这个目标? var articles = ['article10','article9','article8','ar
我有一个刚从文件中读入的 float 的一维 vector 。 std::vector result(s.size() / sizeof(float)); 我想像这样使用这些数据 myTable[rl
使用 numpy 将 data reshape 为 fencepost 的最有效方法是什么? data = np.array([1, 2, 3, 4, 5])
我正在尝试通过删除 for 循环并仅在处理大型数据集时使用 numpy 数组来优化一些代码。 我想采用一维 numpy 数组,例如: a = [1, 2, 3, 4, 5] 并生成一个 2D nump
我正在尝试为两个数组 Pages 和 Price 赋值。 #include int main() { static int pages[3]; static int price[3];
我是深度学习、keras API 和卷积网络的新手,如果这些错误是幼稚的,请事先致歉。我正在尝试构建一个用于分类的简单卷积神经网络。输入数据 X 有 286 个样本,每个样本有 500 个时间点,4
假设我有 Numpy 数组 p 和一个 Scipy 稀疏矩阵 q 这样 >>> p.shape (10,) >>> q.shape (10,100) 我想做 p 和 q 的点积。当我尝试使用 nump
我需要一些行为类似于 std::vector 的东西(界面/功能/等)但我需要它是平坦的,即它不能动态分配缓冲区。显然,这通常不起作用,因为可用大小必须在编译时确定。但我希望类型能够处理N没有额外分配
作为我正在运行的一些模拟的一部分,我需要最终对一些非常长的(实数)数字序列执行以下操作。这是要点: 给定一个长的一维 NumPy 数组,对于数组中的每个位置,我想对该位置前后的值进行平均,取平均值之间
这个问题在这里已经有了答案: Concatenating two one-dimensional NumPy arrays (6 个答案) 关闭 5 年前。 我想将 numpy 数组存储到另一个 n
我是一名优秀的程序员,十分优秀!