mindspore.numpy
MindSpore Numpy package contains a set of Numpy-like interfaces, which allows developers to build models on MindSpore with similar syntax of Numpy.
MindSpore Numpy operators can be classified into four functional modules: array generation, array operation, logic operation and math operation.
Common imported modules in corresponding API examples are as follows:
import mindspore.numpy as np
Note
MindSpore numpy provides a consistent programming experience with native numpy by assembling the low-level operators. Compared with MindSpore’s function and ops interfaces, it is easier for user to understand and use. However, please notice that to be more compatible with native numpy, the performance of some MindSpore numpy interfaces may be weaker than the corresponding function/ops interfaces. Users can choose which to use as needed.
Array Generation
Array generation operators are used to generate tensors.
Here is an example to generate an array:
import mindspore.numpy as np
import mindspore.ops as ops
input_x = np.array([1, 2, 3], np.float32)
print("input_x =", input_x)
print("type of input_x =", ops.typeof(input_x))
The result is as follows:
input_x = [1. 2. 3.]
type of input_x = Tensor[Float32]
Here we have more examples:
Generate a tensor filled with the same element
np.full can be used to generate a tensor with user-specified values:
input_x = np.full((2, 3), 6, np.float32) print(input_x)
The result is as follows:
[[6. 6. 6.] [6. 6. 6.]]
Here is another example to generate an array with the specified shape and filled with the value of 1:
input_x = np.ones((2, 3), np.float32) print(input_x)
The result is as follows:
[[1. 1. 1.] [1. 1. 1.]]
Generate tensors in a specified range
Generate an arithmetic array within the specified range：
input_x = np.arange(0, 5, 1) print(input_x)
The result is as follows:
[0 1 2 3 4]
Generate tensors with specific requirement
Generate a matrix where the lower elements are 1 and the upper elements are 0 on the given diagonal:
input_x = np.tri(3, 3, 1) print(input_x)
The result is as follows:
[[1. 1. 0.] [1. 1. 1.] [1. 1. 1.]]
Another example, generate a 2-D matrix with a diagonal of 1 and other elements of 0:
input_x = np.eye(2, 2) print(input_x)
The result is as follows:
[[1. 0.] [0. 1.]]
API Name |
Description |
Supported Platforms |
Returns evenly spaced values within a given interval. |
|
|
Creates a tensor. |
|
|
Converts the input to tensor. |
|
|
Similar to asarray, converts the input to a float tensor. |
|
|
Returns the Bartlett window. |
|
|
Returns the Blackman window. |
|
|
Returns a tensor copy of the given object. |
|
|
Extracts a diagonal or construct a diagonal array. |
|
|
Returns the indices to access the main diagonal of an array. |
|
|
Creates a two-dimensional array with the flattened input as a diagonal. |
|
|
Returns specified diagonals. |
|
|
Returns a new array of given shape and type, without initializing entries. |
|
|
Returns a new array with the same shape and type as a given array. |
|
|
Returns a 2-D tensor with ones on the diagonal and zeros elsewhere. |
|
|
Returns a new tensor of given shape and type, filled with fill_value. |
|
|
Returns a full array with the same shape and type as a given array. |
|
|
Returns numbers spaced evenly on a log scale (a geometric progression). |
|
|
Returns the Hamming window. |
|
|
Returns the Hanning window. |
|
|
Function to calculate only the edges of the bins used by the histogram function. |
|
|
Returns the identity tensor. |
|
|
Returns an array representing the indices of a grid. |
|
|
Constructs an open mesh from multiple sequences. |
|
|
Returns evenly spaced values within a given interval. |
|
|
Returns numbers spaced evenly on a log scale. |
|
|
Returns coordinate matrices from coordinate vectors. |
|
|
mgrid is an |
|
|
ogrid is an |
|
|
Returns a new tensor of given shape and type, filled with ones. |
|
|
Returns an array of ones with the same shape and type as a given array. |
|
|
Pads an array. |
|
|
Returns a new Tensor with given shape and dtype, filled with random numbers from the uniform distribution on the interval \([0, 1)\). |
|
|
Return random integers from minval (inclusive) to maxval (exclusive). |
|
|
Returns a new Tensor with given shape and dtype, filled with a sample (or samples) from the standard normal distribution. |
|
|
Returns the sum along diagonals of the array. |
|
|
Returns a tensor with ones at and below the given diagonal and zeros elsewhere. |
|
|
Returns a lower triangle of a tensor. |
|
|
Returns the indices for the lower-triangle of an (n, m) array. |
|
|
Returns the indices for the lower-triangle of arr. |
|
|
Returns an upper triangle of a tensor. |
|
|
Returns the indices for the upper-triangle of an (n, m) array. |
|
|
Returns the indices for the upper-triangle of arr. |
|
|
Generates a Vandermonde matrix. |
|
|
Returns a new tensor of given shape and type, filled with zeros. |
|
|
Returns an array of zeros with the same shape and type as a given array. |
|
Array Operation
Array operations focus on tensor manipulation.
Manipulate the shape of the tensor
For example, transpose a matrix:
input_x = np.arange(10).reshape(5, 2) output = np.transpose(input_x) print(output)
The result is as follows:
[[0 2 4 6 8] [1 3 5 7 9]]
Another example, swap two axes:
input_x = np.ones((1, 2, 3)) output = np.swapaxes(input_x, 0, 1) print(output.shape)
The result is as follows:
(2, 1, 3)
Tensor splitting
Divide the input tensor into multiple tensors equally, for example:
input_x = np.arange(9) output = np.split(input_x, 3) print(output)
The result is as follows:
(Tensor(shape=[3], dtype=Int32, value= [0, 1, 2]), Tensor(shape=[3], dtype=Int32, value= [3, 4, 5]), Tensor(shape=[3], dtype=Int32, value= [6, 7, 8]))
Tensor combination
Concatenate the two tensors according to the specified axis, for example:
input_x = np.arange(0, 5) input_y = np.arange(10, 15) output = np.concatenate((input_x, input_y), axis=0) print(output)
The result is as follows:
[ 0 1 2 3 4 10 11 12 13 14]
API Name |
Description |
Supported Platforms |
Appends values to the end of a tensor. |
|
|
Applies a function to 1-D slices along the given axis. |
|
|
Applies a function repeatedly over multiple axes. |
|
|
Splits a tensor into multiple sub-tensors. |
|
|
Returns a string representation of the data in an array. |
|
|
Converts inputs to arrays with at least one dimension. |
|
|
Reshapes inputs as arrays with at least two dimensions. |
|
|
Reshapes inputs as arrays with at least three dimensions. |
|
|
Broadcasts any number of arrays against each other. |
|
|
Broadcasts an array to a new shape. |
|
|
Construct an array from an index array and a list of arrays to choose from. |
|
|
Stacks 1-D tensors as columns into a 2-D tensor. |
|
|
Joins a sequence of tensors along an existing axis. |
|
|
Splits a tensor into multiple sub-tensors along the 3rd axis (depth). |
|
|
Stacks tensors in sequence depth wise (along the third axis). |
|
|
Expands the shape of a tensor. |
|
|
Reverses the order of elements in an array along the given axis. |
|
|
Flips the entries in each row in the left/right direction. |
|
|
Flips the entries in each column in the up/down direction. |
|
|
Splits a tensor into multiple sub-tensors horizontally (column-wise). |
|
|
Stacks tensors in sequence horizontally. |
|
|
Moves axes of an array to new positions. |
|
|
Evaluates a piecewise-defined function. |
|
|
Returns a contiguous flattened tensor. |
|
|
Repeats elements of an array. |
|
|
Reshapes a tensor without changing its data. |
|
|
Rolls a tensor along given axes. |
|
|
Rolls the specified axis backwards, until it lies in the given position. |
|
|
Rotates a tensor by 90 degrees in the plane specified by axes. |
|
|
Returns an array drawn from elements in choicelist, depending on conditions. |
|
|
Returns the number of elements along a given axis. |
|
|
Splits a tensor into multiple sub-tensors along the given axis. |
|
|
Removes single-dimensional entries from the shape of a tensor. |
|
|
Joins a sequence of arrays along a new axis. |
|
|
Interchanges two axes of a tensor. |
|
|
Takes elements from an array along an axis. |
|
|
Takes values from the input array by matching 1d index and data slices. |
|
|
Constructs an array by repeating a the number of times given by reps. |
|
|
Reverses or permutes the axes of a tensor; returns the modified tensor. |
|
|
Finds the unique elements of a tensor. |
|
|
Converts a flat index or array of flat indices into a tuple of coordinate arrays. |
|
|
Splits a tensor into multiple sub-tensors vertically (row-wise). |
|
|
Stacks tensors in sequence vertically. |
|
|
Returns elements chosen from x or y depending on condition. |
|
Logic
Logic operations define computations related with boolean types. Examples of equal and less operations are as follows:
input_x = np.arange(0, 5)
input_y = np.arange(0, 10, 2)
output = np.equal(input_x, input_y)
print("output of equal:", output)
output = np.less(input_x, input_y)
print("output of less:", output)
The result is as follows:
output of equal: [ True False False False False]
output of less: [False True True True True]
API Name |
Description |
Supported Platforms |
Returns True if input arrays have same shapes and all elements equal. |
|
|
Returns True if input arrays are shape consistent and all elements equal. |
|
|
Returns the truth value of |
|
|
Returns the truth value of |
|
|
Returns the truth value of |
|
|
Tests whether each element of a 1-D array is also present in a second array. |
|
|
Returns a boolean tensor where two tensors are element-wise equal within a tolerance. |
|
|
Tests element-wise for finiteness (not infinity or not Not a Number). |
|
|
Calculates element in test_elements, broadcasting over element only. |
|
|
Tests element-wise for positive or negative infinity. |
|
|
Tests element-wise for NaN and return result as a boolean array. |
|
|
Tests element-wise for negative infinity, returns result as bool array. |
|
|
Tests element-wise for positive infinity, returns result as bool array. |
|
|
Returns True if the type of element is a scalar type. |
|
|
Returns the truth value of |
|
|
Returns the truth value of |
|
|
Computes the truth value of x1 AND x2 element-wise. |
|
|
Computes the truth value of NOT a element-wise. |
|
|
Computes the truth value of x1 OR x2 element-wise. |
|
|
Computes the truth value of x1 XOR x2, element-wise. |
|
|
Returns (x1 != x2) element-wise. |
|
|
Returns element-wise True where signbit is set (less than zero). |
|
|
Tests whether any array element along a given axis evaluates to True. |
|
Math
Math operations include basic and advanced math operations on tensors, and they have full support on Numpy broadcasting rules. Here are some examples:
Sum two tensors
The following code implements the operation of adding two tensors of input_x and input_y:
input_x = np.full((3, 2), [1, 2]) input_y = np.full((3, 2), [3, 4]) output = np.add(input_x, input_y) print(output)
The result is as follows:
[[4 6] [4 6] [4 6]]
Matrics multiplication
The following code implements the operation of multiplying two matrices input_x and input_y:
input_x = np.arange(2*3).reshape(2, 3).astype('float32') input_y = np.arange(3*4).reshape(3, 4).astype('float32') output = np.matmul(input_x, input_y) print(output)
The result is as follows:
[[20. 23. 26. 29.] [56. 68. 80. 92.]]
Take the average along a given axis
The following code implements the operation of averaging all the elements of input_x:
input_x = np.arange(6).astype('float32') output = np.mean(input_x) print(output)
The result is as follows:
2.5Exponential arithmetic
The following code implements the operation of the natural constant e to the power of input_x:
input_x = np.arange(5).astype('float32') output = np.exp(input_x) print(output)
The result is as follows:
[ 1. 2.7182817 7.389056 20.085537 54.59815 ]
API Name |
Description |
Supported Platforms |
Calculates the absolute value element-wise. |
|
|
Adds arguments element-wise. |
|
|
Returns the maximum of an array or maximum along an axis. |
|
|
Returns the minimum of an array or minimum along an axis. |
|
|
Trigonometric inverse cosine, element-wise. |
|
|
Inverse hyperbolic cosine, element-wise. |
|
|
Inverse sine, element-wise. |
|
|
Inverse hyperbolic sine element-wise. |
|
|
Trigonometric inverse tangent, element-wise. |
|
|
Element-wise arc tangent of \(x1/x2\) choosing the quadrant correctly. |
|
|
Inverse hyperbolic tangent element-wise. |
|
|
Returns the indices of the maximum values along an axis. |
|
|
Returns the indices of the minimum values along an axis. |
|
|
Evenly round to the given number of decimals. |
|
|
Computes the weighted average along the specified axis. |
|
|
Count number of occurrences of each value in array of non-negative ints. |
|
|
Computes the bit-wise AND of two arrays element-wise. |
|
|
Computes the bit-wise OR of two arrays element-wise. |
|
|
Computes the bit-wise XOR of two arrays element-wise. |
|
|
Returns the cube-root of a tensor, element-wise. |
|
|
Returns the ceiling of the input, element-wise. |
|
|
Clips (limits) the values in an array. |
|
|
Returns the discrete, linear convolution of two one-dimensional sequences. |
|
|
Changes the sign of x1 to that of x2, element-wise. |
|
|
Returns Pearson product-moment correlation coefficients. |
|
|
Cross-correlation of two 1-dimensional sequences. |
|
|
Cosine element-wise. |
|
|
Hyperbolic cosine, element-wise. |
|
|
Counts the number of non-zero values in the tensor x. |
|
|
Estimates a covariance matrix, given data and weights. |
|
|
Returns the cross product of two (arrays of) vectors. |
|
|
Returns the cumulative product of elements along a given axis. |
|
|
Returns the cumulative sum of the elements along a given axis. |
|
|
Converts angles from degrees to radians. |
|
|
Calculates the n-th discrete difference along the given axis. |
|
|
Returns the indices of the bins to which each value in input array belongs. |
|
|
Returns a true division of the inputs, element-wise. |
|
|
Returns element-wise quotient and remainder simultaneously. |
|
|
Returns the dot product of two arrays. |
|
|
The differences between consecutive elements of a tensor. |
|
|
Calculates the exponential of all elements in the input array. |
|
|
Calculates |
|
|
Calculates |
|
|
Rounds to nearest integer towards zero. |
|
|
First array elements raised to powers from second array, element-wise. |
|
|
Returns the floor of the input, element-wise. |
|
|
Returns the largest integer smaller or equal to the division of the inputs. |
|
|
Returns the element-wise remainder of division. |
|
|
Returns the greatest common divisor of |
|
|
Returns the gradient of a N-dimensional array. |
|
|
Computes the Heaviside step function. |
|
|
Computes the histogram of a dataset. |
|
|
Computes the multidimensional histogram of some data. |
|
|
Computes the multidimensional histogram of some data. |
|
|
Given the "legs" of a right triangle, returns its hypotenuse. |
|
|
Returns the inner product of two tensors. |
|
|
One-dimensional linear interpolation for monotonically increasing sample points. |
|
|
Computes bit-wise inversion, or bit-wise NOT, element-wise. |
|
|
Kronecker product of two arrays. |
|
|
Returns the lowest common multiple of |
|
|
Returns the natural logarithm, element-wise. |
|
|
Base-10 logarithm of x. |
|
|
Returns the natural logarithm of one plus the input array, element-wise. |
|
|
Base-2 logarithm of x. |
|
|
Logarithm of the sum of exponentiations of the inputs. |
|
|
Logarithm of the sum of exponentiations of the inputs in base of 2. |
|
|
Returns the matrix product of two arrays. |
|
|
Raises a square matrix to the (integer) power n. |
|
|
Returns the element-wise maximum of array elements. |
|
|
Computes the arithmetic mean along the specified axis. |
|
|
Element-wise minimum of tensor elements. |
|
|
Computes the dot product of two or more arrays in a single function call, while automatically selecting the fastest evaluation order. |
|
|
Multiplies arguments element-wise. |
|
|
Return the cumulative sum of array elements over a given axis treating Not a Numbers (NaNs) as zero. |
|
|
Return the maximum of an array or maximum along an axis, ignoring any NaNs. |
|
|
Computes the arithmetic mean along the specified axis, ignoring NaNs. |
|
|
Returns the minimum of array elements over a given axis, ignoring any NaNs. |
|
|
Computes the standard deviation along the specified axis, while ignoring NaNs. |
|
|
Returns the sum of array elements over a given axis treating Not a Numbers (NaNs) as zero. |
|
|
Computes the variance along the specified axis, while ignoring NaNs. |
|
|
Numerical negative, element-wise. |
|
|
Matrix or vector norm. |
|
|
Computes the outer product of two vectors. |
|
|
Finds the sum of two polynomials. |
|
|
Returns the derivative of the specified order of a polynomial. |
|
|
Returns an antiderivative (indefinite integral) of a polynomial. |
|
|
Finds the product of two polynomials. |
|
|
Difference (subtraction) of two polynomials. |
|
|
Evaluates a polynomial at specific values. |
|
|
Numerical positive, element-wise. |
|
|
First array elements raised to powers from second array, element-wise. |
|
|
Returns the data type with the smallest size and smallest scalar kind. |
|
|
Range of values (maximum - minimum) along an axis. |
|
|
Converts angles from radians to degrees. |
|
|
Converts angles from degrees to radians. |
|
|
Converts a tuple of index arrays into an array of flat indices, applying boundary modes to the multi-index. |
|
|
Returns the reciprocal of the argument, element-wise. |
|
|
Returns element-wise remainder of division. |
|
|
Returns the type that results from applying the type promotion rules to the arguments. |
|
|
Rounds elements of the array to the nearest integer. |
|
|
Finds indices where elements should be inserted to maintain order. |
|
|
Returns an element-wise indication of the sign of a number. |
|
|
Trigonometric sine, element-wise. |
|
|
Hyperbolic sine, element-wise. |
|
|
Returns the non-negative square-root of an array, element-wise. |
|
|
Returns the element-wise square of the input. |
|
|
Computes the standard deviation along the specified axis. |
|
|
Subtracts arguments, element-wise. |
|
|
Returns sum of array elements over a given axis. |
|
|
Computes tangent element-wise. |
|
|
Computes hyperbolic tangent element-wise. |
|
|
Computes tensor dot product along specified axes. |
|
|
Integrates along the given axis using the composite trapezoidal rule. |
|
|
Returns a true division of the inputs, element-wise. |
|
|
Returns the truncated value of the input, element-wise. |
|
|
Unwraps by changing deltas between values to |
|
|
Computes the variance along the specified axis. |
|
Interact With MindSpore Functions
Since mindspore.numpy directly wraps MindSpore tensors and operators, it has all the advantages and properties of MindSpore. In this section, we will briefly introduce how to employ MindSpore execution management and automatic differentiation in mindspore.numpy coding scenarios. These include:
ms_function: for running codes in static graph mode for better efficiency.
GradOperation: for automatic gradient computation.
mindspore.set_context: for mindspore.numpy execution management.
mindspore.nn.Cell: for using mindspore.numpy interfaces in MindSpore Deep Learning Models.
The following are examples:
Use ms_function to run code in static graph mode
Let’s first see an example consisted of matrix multiplication and bias add, which is a typical process in Neural Networks:
import mindspore.numpy as np x = np.arange(8).reshape(2, 4).astype('float32') w1 = np.ones((4, 8)) b1 = np.zeros((8,)) w2 = np.ones((8, 16)) b2 = np.zeros((16,)) w3 = np.ones((16, 4)) b3 = np.zeros((4,)) def forward(x, w1, b1, w2, b2, w3, b3): x = np.dot(x, w1) + b1 x = np.dot(x, w2) + b2 x = np.dot(x, w3) + b3 return x print(forward(x, w1, b1, w2, b2, w3, b3))
The result is as follows:
[[ 768. 768. 768. 768.] [2816. 2816. 2816. 2816.]]
In this function, MindSpore dispatches each computing kernel to device separately. However, with the help of ms_function, we can compile all operations into a single static computing graph.
from mindspore import ms_function forward_compiled = ms_function(forward) print(forward(x, w1, b1, w2, b2, w3, b3))
The result is as follows:
[[ 768. 768. 768. 768.] [2816. 2816. 2816. 2816.]]
Note
Currently, static graph cannot run in Python interactive mode and not all python types can be passed into functions decorated with ms_function. For details about how to use ms_function, see API ms_function .
Use GradOperation to compute deratives
GradOperation can be used to take deratives from normal functions and functions decorated with ms_function. Take the previous example:
from mindspore import ops grad_all = ops.composite.GradOperation(get_all=True) print(grad_all(forward)(x, w1, b1, w2, b2, w3, b3))
The result is as follows:
(Tensor(shape=[2, 4], dtype=Float32, value= [[ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02, 5.12000000e+02], [ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02, 5.12000000e+02]]), Tensor(shape=[4, 8], dtype=Float32, value= [[ 2.56000000e+02, 2.56000000e+02, 2.56000000e+02 ... 2.56000000e+02, 2.56000000e+02, 2.56000000e+02], [ 3.84000000e+02, 3.84000000e+02, 3.84000000e+02 ... 3.84000000e+02, 3.84000000e+02, 3.84000000e+02], [ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02 ... 5.12000000e+02, 5.12000000e+02, 5.12000000e+02] [ 6.40000000e+02, 6.40000000e+02, 6.40000000e+02 ... 6.40000000e+02, 6.40000000e+02, 6.40000000e+02]]), ... Tensor(shape=[4], dtype=Float32, value= [ 2.00000000e+00, 2.00000000e+00, 2.00000000e+00, 2.00000000e+00]))
To take the gradient of ms_function compiled functions, first we need to set the execution mode to static graph mode.
from mindspore import ms_function, set_context, GRAPH_MODE set_context(mode=GRAPH_MODE) grad_all = ops.composite.GradOperation(get_all=True) print(grad_all(ms_function(forward))(x, w1, b1, w2, b2, w3, b3))
The result is as follows:
(Tensor(shape=[2, 4], dtype=Float32, value= [[ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02, 5.12000000e+02], [ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02, 5.12000000e+02]]), Tensor(shape=[4, 8], dtype=Float32, value= [[ 2.56000000e+02, 2.56000000e+02, 2.56000000e+02 ... 2.56000000e+02, 2.56000000e+02, 2.56000000e+02], [ 3.84000000e+02, 3.84000000e+02, 3.84000000e+02 ... 3.84000000e+02, 3.84000000e+02, 3.84000000e+02], [ 5.12000000e+02, 5.12000000e+02, 5.12000000e+02 ... 5.12000000e+02, 5.12000000e+02, 5.12000000e+02] [ 6.40000000e+02, 6.40000000e+02, 6.40000000e+02 ... 6.40000000e+02, 6.40000000e+02, 6.40000000e+02]]), ... Tensor(shape=[4], dtype=Float32, value= [ 2.00000000e+00, 2.00000000e+00, 2.00000000e+00, 2.00000000e+00]))
For more details, see API GradOperation .
Use mindspore.set_context to control execution mode
Most functions in mindspore.numpy can run in Graph Mode and PyNative Mode, and can run on CPU, GPU and Ascend. Like MindSpore, users can manage the execution mode using mindspore.set_context：
from mindspore import set_context, GRAPH_MODE, PYNATIVE_MODE # Execucation in static graph mode set_context(mode=GRAPH_MODE) # Execucation in PyNative mode set_context(mode=PYNATIVE_MODE) # Execucation on CPU backend set_context(device_target="CPU") # Execucation on GPU backend set_context(device_target="GPU") # Execucation on Ascend backend set_context(device_target="Ascend") ...
For more details, see API mindspore.set_context .
Use mindspore.numpy in MindSpore Deep Learning Models
mindspore.numpy interfaces can be used inside nn.cell blocks as well. For example, the above code can be modified to:
import mindspore.numpy as np from mindspore import set_context, GRAPH_MODE from mindspore.nn import Cell set_context(mode=GRAPH_MODE) x = np.arange(8).reshape(2, 4).astype('float32') w1 = np.ones((4, 8)) b1 = np.zeros((8,)) w2 = np.ones((8, 16)) b2 = np.zeros((16,)) w3 = np.ones((16, 4)) b3 = np.zeros((4,)) class NeuralNetwork(Cell): def construct(self, x, w1, b1, w2, b2, w3, b3): x = np.dot(x, w1) + b1 x = np.dot(x, w2) + b2 x = np.dot(x, w3) + b3 return x net = NeuralNetwork() print(net(x, w1, b1, w2, b2, w3, b3))
The result is as follows:
[[ 768. 768. 768. 768.] [2816. 2816. 2816. 2816.]]