NDArray Basics using MXNet

NDArray module is a primary tool of MXNet, it is used for storing and transforming the data. It is just like NumPy's multi-dimensional array. It has some advantages like, NDArrays support asynchronous computation on CPU, GPU and distributed cloud architectures. NDArrays provide support for automatic differentiation. So these advantages make the NDArray indispensable for deep learning.

NDArrays are multi-dimensional arrays of numerical values. NDArrays with one axis corresponds to vectors, two axes to matrices, more than two axes it corresponds to tensors.

to use mxnet in python, you need to install in your PC by typing at command prompt as shown below:
C:\Users\xxxx> pip install mxnet

to get started, let's import mxnet and import ndarray from mxnet.

>>> import mxnet as mx
>>> from mxnet import nd

(1). we can create a simple 1-dimensional array using mxnet from a python list, like this :
>>> x = nd.array([1, 2, 3])
>>> print(x)
[1. 2. 3.]
<NDArray 3 @cpu(0)>

<NDArray 3 @cpu(0)> indicates that x is a one-dimensional array of length 3 and it resides in CPU main memory. The 0 in @cpu(0) has no special meaning and does not represent a specific core.

(2). we can create a 2-dimensional array using mxnet from a python list, like this :
>>> y = nd.array([[1, 2, 3, 4], [1, 2, 3, 4], [1, 2, 3, 4]])
>>> print(y)
[[1. 2. 3. 4.]
 [1. 2. 3. 4.]
 [1. 2. 3. 4.]]
<NDArray 3x4 @cpu(0)>

(3). we can create an empty 2D array ( also called matrix) with 3 rows and 3 columns like this :
>>> x = nd.empty((3, 3))
>>> print(x)
[[0. 0. 0.]
 [0. 0. 0.]
 [0. 0. 0.]]
<NDArray 3x3 @cpu(0)>
if empty function is used, it grabs some memory and gives us back a matrix without setting the values of any of its entries. This means that the entries can have any form of values.

(4). if we want our matrices to be initialized with zeros, then we have to use .zeros function like this:
>>> x = nd.zeros((3, 3))
>>> print(x)
[[0. 0. 0.]
 [0. 0. 0.]
 [0. 0. 0.]]
<NDArray 3x3 @cpu(0)>

(5). similarly, ndarray has a function to create a matrix of all ones, use .ones function like this:
>>> x = nd.ones((3, 4))
>>> print(x)
[[1. 1. 1. 1.]
 [1. 1. 1. 1.]
 [1. 1. 1. 1.]]
<NDArray 3x4 @cpu(0)>


(6). we can fill with a value ( for Example 7)in a 2D array with 3 rows and 3 columns like this:
>>> x = nd.full((3, 3), 7)
>>> print(x)
[[7. 7. 7.]
 [7. 7. 7.]
 [7. 7. 7.]]
<NDArray 3x3 @cpu(0)>

(7). sometimes, we need to create an array of random values ( this is very common in neural networks ) to use the array as a parameter. For that we can use random_normal function with a zero mean and unit variance form standard normal distribution like this :
>>> y = nd.random_normal(0, 1, shape=(3,4))
>>> print(y)
[[ 1.1630785   0.4838046   0.29956347  0.15302546]
 [-1.1688148   1.558071   -0.5459446  -2.3556297 ]
 [ 0.54144025  2.6785064   1.2546344  -0.54877406]]
<NDArray 3x4 @cpu(0)>

(8). sometimes, you need to copy an array by its shape but not its contents, then use .zeros_like( ) function like this:
>>> z = nd.zeros_like(y)
>>> print(z)
[[0. 0. 0. 0.]
 [0. 0. 0. 0.]
 [0. 0. 0. 0.]]
<NDArray 3x4 @cpu(0)>

(9). you can access the dimensions of array using .shape attribute, like this:
>>> y.shape
(3, 4)

(10). you can access the size of array using .size attribute, like this:
>>> y.size
12

(11). you can query the data type using .dtype, like this:
>>> y.dtype
numpy.float32

float32 is the default data type.

(12). Operations and memory storage of your device can be revealed by using .context attribute, like this:
>>> y.context
cpu(0)