Gluon学习03-基础数据类型Ndarray

Gluon学习03-基础数据类型Ndarray

小书匠 kindle 

目录,方便快速定位:

• 1.安装cuda与cudnn
• 2.安装mxnet-gpu

---

本机环境介绍:

系统:Linuxmint
Python版本:Python3

---

1.API介绍

MxNet版本:1.2.0
API地址:https://mxnet.incubator.apache.org/api/python/ndarray/ndarray.html

Ndarray在CPU/GPU上提供必要的张量操作,是一个多维的,固定大小的,同类型的矩阵.mxnet.ndarray与numpy.ndarray非常相似.

The NDArray class:
1.属性
shape/size/ndim/context/dtype

1. import mxnet as mx 

   ---

2. from mxnet import nd 

   ---

3. import numpy as np 

   ---

4. #数据的形状 

   ---

5. >>> x = mx.nd.array([1, 2, 3, 4]) 

   ---

6. >>> x.shape 

   ---

7. (4L,) 

   ---

8. >>> y = mx.nd.zeros((2, 3, 4)) 

   ---

9. >>> y.shape 

   ---

10. (2L, 3L, 4L) 

    ---

11.  

    ---

12. #数据的多少 

    ---

13. >>> import numpy as np 

    ---

14. >>> x = mx.nd.zeros((3, 5, 2)) 

    ---

15. >>> x.size 

    ---

16. 30 

    ---

17. >>> np.prod(x.shape) 

    ---

18. 30 

    ---

19.  

    ---

20. #数据的阶/秩 

    ---

21. >>> x = mx.nd.array([1, 2, 3, 4]) 

    ---

22. >>> x.ndim 

    ---

23. 1 

    ---

24. >>> x = mx.nd.array([[1, 2], [3, 4]]) 

    ---

25. >>> x.ndim 

    ---

26. 2 

    ---

27.  

    ---

28. #数据所在的设备 

    ---

29. >>> x = mx.nd.array([1, 2, 3, 4]) 

    ---

30. >>> x.context 

    ---

31. cpu(0) 

    ---

32. >>> type(x.context) 

    ---

33.  

    ---

34. >>> y = mx.nd.zeros((2,3), mx.gpu(0)) 

    ---

35. >>> y.context 

    ---

36. gpu(0) 

    ---

37.  

    ---

38. #数据的类型 

    ---

39. >>> x = mx.nd.zeros((2,3)) 

    ---

40. >>> x.dtype 

    ---

41.  

    ---

42. >>> y = mx.nd.zeros((2,3), dtype='int32') 

    ---

43. >>> y.dtype 

    ---

2.转换

1. #转为标量,形状必须是(1,) 

   ---

2. >>> x = mx.nd.ones((1,), dtype='int32') 

   ---

3. >>> x.asscalar() 

   ---

4. 1 

   ---

5. >>> type(x.asscalar()) 

   ---

6.  

   ---

7. #复制 

   ---

8. >>> x = mx.nd.ones((2,3)) 

   ---

9. >>> y = x.copy() 

   ---

10. >>> y.asnumpy() 

    ---

11. array([[ 1., 1., 1.], 

    ---

12. [ 1., 1., 1.]], dtype=float32) 

    ---

3.创建

1. #通过自身API创建 

   ---

2. >>> a=nd.arange((10)) 

   ---

3. >>> a 

   ---

4. [0. 1. 2. 3. 4. 5. 6. 7. 8. 9.] 

   ---

5. <NDArray 10 @cpu(0)> 

   ---

6.  

   ---

7. >>> b=nd.zeros((2,3)) 

   ---

8. >>> b 

   ---

9. [[0. 0. 0.] 

   ---

10. [0. 0. 0.]] 

    ---

11. <NDArray 2x3 @cpu(0)> 

    ---

12.  

    ---

13. >>> c=nd.ones((2,3,1)) 

    ---

14. >>> c 

    ---

15. [[[1.] 

    ---

16. [1.] 

    ---

17. [1.]] 

    ---

18. [[1.] 

    ---

19. [1.] 

    ---

20. [1.]]] 

    ---

21. <NDArray 2x3x1 @cpu(0)> 

    ---

22.  

    ---

23. #通过list创建 

    ---

24. >>> d=[6,5,4,3,2,1] 

    ---

25. >>> e=nd.array(d) 

    ---

26. >>> e 

    ---

27. [6. 5. 4. 3. 2. 1.] 

    ---

28. <NDArray 6 @cpu(0)> 

    ---

4.形状

1. #转置 

   ---

2. >>> x = mx.nd.arange(0,6).reshape((2,3)) 

   ---

3. >>> x.asnumpy() 

   ---

4. array([[ 0., 1., 2.], 

   ---

5. [ 3., 4., 5.]], dtype=float32) 

   ---

6. >>> x.T.asnumpy() 

   ---

7. array([[ 0., 3.], 

   ---

8. [ 1., 4.], 

   ---

9. [ 2., 5.]], dtype=float32) 

   ---

10.  

    ---

11. #改变形状 

    ---

12. >>> x = mx.nd.arange(0,6).reshape(2,3) 

    ---

13. >>> x 

    ---

14. [[0. 1. 2.] 

    ---

15. [3. 4. 5.]] 

    ---

16. <NDArray 2x3 @cpu(0)> 

    ---

17.  

    ---

18. >>> y = x.reshape(3,2) 

    ---

19. >>> y 

    ---

20. [[0. 1.] 

    ---

21. [2. 3.] 

    ---

22. [4. 5.]] 

    ---

23. <NDArray 3x2 @cpu(0)> 

    ---

24.  

    ---

25. #列多少不管,就明确是n行,列= (x.size/n)上整 

    ---

26. >>> y = x.reshape(2,-1) 

    ---

27. >>> y 

    ---

28. [[0. 1. 2.] 

    ---

29. [3. 4. 5.]] 

    ---

30. <NDArray 2x3 @cpu(0)> 

    ---

31.  

    ---

32. #只要一行 

    ---

33. >>> y = x.reshape(-3) 

    ---

34. >>> y 

    ---

35. [0. 1. 2. 3. 4. 5.] 

    ---

36. <NDArray 6 @cpu(0)> 

    ---

37. >>>  

    ---

38.  

    ---

5.元素扩展

1. #拼接,输入数组的唯独应该相同 

   ---

2. x = [[1,1],[2,2]] 

   ---

3. y = [[3,3],[4,4],[5,5]] 

   ---

4. z = [[6,6], [7,7],[8,8]] 

   ---

5. concat(x,y,z,dim=0) = [[ 1., 1.], 

   ---

6. [ 2., 2.], 

   ---

7. [ 3., 3.], 

   ---

8. [ 4., 4.], 

   ---

9. [ 5., 5.], 

   ---

10. [ 6., 6.], 

    ---

11. [ 7., 7.], 

    ---

12. [ 8., 8.]] 

    ---

参考:
NDArray API

每日一学,争取进步03