Skip to content

jjjkkkjjj/Matft

BranchesTags
 
 

Repository files navigation

Matft

SwiftPM compatible CocoaPods compatible Carthage compatible license

Matft is Numpy-like library in Swift. Function name and usage is similar to Numpy.

Protocol Version(Beta)

This version is type-safe, but slightly slower than original one(master branch)

Feature & Usage

  • Many types

  • Pretty print

  • Indexing

    • Positive
    • Negative
    • Boolean

  • Slicing

    • Start / To / By
    • New Axis

  • View

    • Assignment

  • Conversion

    • Broadcast
    • Transpose
    • Reshape
    • Astype

  • Mathematic

    • Arithmetic
    • Statistic
    • Linear Algebra

...etc.

See Function List for all functions.

Declaration

MfArray

  • The MfArray such like a numpy.ndarray

    let a = MfArray<Int>([[[ -8,  -7,  -6,  -5],
                  [ -4,  -3,  -2,  -1]],
        
                 [[ 0,  1,  2,  3],
                  [ 4,  5,  6,  7]]])
let aa = Matft.arange(start: -8, to: 8, by: 1, shape: [2,2,4])
print(a)
print(aa)
/*
mfarray = 
[[[	-8,		-7,		-6,		-5],
[	-4,		-3,		-2,		-1]],

[[	0,		1,		2,		3],
[	4,		5,		6,		7]]], type=Int, shape=[2, 2, 4]
mfarray = 
[[[	-8,		-7,		-6,		-5],
[	-4,		-3,		-2,		-1]],

[[	0,		1,		2,		3],
[	4,		5,		6,		7]]], type=Int, shape=[2, 2, 4]
*/

MfTypable

  • You can use specific Type comformed to MfTypable.

    ※Note that stored data type will be Float or Double only (comformed MfStorable) even if you set Int. So, if you input big number to MfArray, it may be cause to overflow or strange results in any calculation (+, -, *, /,... etc.). But I believe this is not problem in practical use.

  • MfTypeable list is below

      Bool, UInt8, UInt16, UInt32, UInt64, UInt, Int8, Int16, Int32, Int64, Int, Float, Double: MfTypable

  • Also, you can convert MfType easily using astype

    let a = MfArray<Int>([[[ -8,  -7,  -6,  -5],
                  [ -4,  -3,  -2,  -1]],
        
                 [[ 0,  1,  2,  3],
                  [ 4,  5,  6,  7]]])

print(aa.astype(Double.self))
/*
mfarray = 
[[[	-8.0,		-7.0,		-6.0,		-5.0],
[	-4.0,		-3.0,		-2.0,		-1.0]],

[[	0.0,		1.0,		2.0,		3.0],
[	4.0,		5.0,		6.0,		7.0]]], type=Double, shape=[2, 2, 4]
*/

Subscription

MfSlice

  • You can access specific data using subscript.

You can set MfSlice (see below's list) to subscript.

  • MfSlice(start: Int? = nil, to: Int? = nil, by: Int = 1)
  • Matft.newaxis
  • ~< //this is prefix, postfix and infix operator. same as python's slice, ":"

(Positive) Indexing

  • Normal indexing

    let a = Matft.arange(start: 0, to: 27, by: 1, shape: [3,3,3])
print(a)
/*
mfarray = 
[[[	0,		1,		2],
[	3,		4,		5],
[	6,		7,		8]],

[[	9,		10,		11],
[	12,		13,		14],
[	15,		16,		17]],

[[	18,		19,		20],
[	21,		22,		23],
[	24,		25,		26]]], type=Int, shape=[3, 3, 3]
*/
print(a[2,1,0])
// 21

    Slicing

  • If you replace : with ~<, you can get sliced mfarray. Note that use a[0~<] instead of a[:] to get all elements along axis.

    print(a[~<1])  //same as a[:1] for numpy
/*
mfarray = 
[[[	9,		10,		11],
[	12,		13,		14],
[	15,		16,		17]]], type=Int, shape=[1, 3, 3]
*/
print(a[1~<3]) //same as a[1:3] for numpy
/*
mfarray = 
[[[	9,		10,		11],
[	12,		13,		14],
[	15,		16,		17]],

[[	18,		19,		20],
[	21,		22,		23],
[	24,		25,		26]]], type=Int, shape=[2, 3, 3]
*/
print(a[~<~<2]) //same as a[::2] for numpy
//print(a[~<<2]) //alias
/*
mfarray = 
[[[	0,		1,		2],
[	3,		4,		5],
[	6,		7,		8]],

[[	18,		19,		20],
[	21,		22,		23],
[	24,		25,		26]]], type=Int, shape=[2, 3, 3]
*/

Negative Indexing

  • Negative indexing is also available That's implementation was hardest for me...

    print(a[~<-1])
/*
mfarray = 
[[[	0,		1,		2],
[	3,		4,		5],
[	6,		7,		8]],

[[	9,		10,		11],
[	12,		13,		14],
[	15,		16,		17]]], type=Int, shape=[2, 3, 3]
*/
print(a[-1~<-3])
/*
mfarray = 
	[], type=Int, shape=[0, 3, 3]
*/
print(a[~<~<-1])
//print(a[~<<-1]) //alias
/*
mfarray = 
[[[	18,		19,		20],
[	21,		22,		23],
[	24,		25,		26]],

[[	9,		10,		11],
[	12,		13,		14],
[	15,		16,		17]],

[[	0,		1,		2],
[	3,		4,		5],
[	6,		7,		8]]], type=Int, shape=[3, 3, 3]*/

Boolean Indexing(New!!!)

  • You can use boolean indexing.

    Caution! I don't check performance, so this boolean indexing may be slow

    let img = MfArray<UInt8>([[1, 2, 3],
                               [4, 5, 6],
                               [7, 8, 9]])
img[img > 3] = MfArray<UInt8>([10])
print(img)
/*
mfarray = 
[[	1,		2,		3],
[	10,		10,		10],
[	10,		10,		10]], type=UInt8, shape=[3, 3]
*/

View

  • Note that returned subscripted mfarray will have base property (is similar to view in Numpy). See numpy doc in detail.

    let a = Matft.arange(start: 0, to: 4*4*2, by: 1, shape: [4,4,2])
            
let b = a[0~<, 1]
b[~<<-1] = MfArray<Int>([9999]) // cannot pass Int directly such like 9999

print(a)
/*
mfarray = 
[[[	0,		1],
[	9999,		9999],
[	4,		5],
[	6,		7]],

[[	8,		9],
[	9999,		9999],
[	12,		13],
[	14,		15]],

[[	16,		17],
[	9999,		9999],
[	20,		21],
[	22,		23]],

[[	24,		25],
[	9999,		9999],
[	28,		29],
[	30,		31]]], type=Int, shape=[4, 4, 2]
*/

Function List

Below is Matft's function list. As I mentioned above, almost functions are similar to Numpy. Also, these function use Accelerate framework inside, the perfomance may keep high.

* means method function exists too. Shortly, you can use a.shallowcopy() where a is MfArray.

^ means method function only. Shortly, you can use a.tolist() not Matft.tolist where a is MfArray.

  • Creation
Matft Numpy
*Matft.shallowcopy *numpy.copy
*Matft.deepcopy copy.deepcopy
Matft.nums numpy.ones * N
Matft.nums_like numpy.ones_like * N
Matft.arange numpy.arange
Matft.eye numpy.eye
Matft.diag numpy.diag
Matft.vstack numpy.vstack
Matft.hstack numpy.hstack
Matft.concatenate numpy.concatenate
  • Conversion
Matft Numpy
*Matft.astype *numpy.astype
*Matft.transpose *numpy.transpose
*Matft.expand_dims *numpy.expand_dims
*Matft.squeeze *numpy.squeeze
*Matft.broadcast_to *numpy.broadcast_to
*Matft.conv_order *numpy.ascontiguousarray
*Matft.flatten *numpy.flatten
*Matft.flip *numpy.flip
*Matft.clip *numpy.clip
*Matft.swapaxes *numpy.swapaxes
*Matft.moveaxis *numpy.moveaxis
*Matft.sort *numpy.sort
*Matft.argsort *numpy.argsort
^MfArray.toArray ^numpy.ndarray.tolist

  • File

    save function has not developed yet.

Matft Numpy
Matft.file.loadtxt numpy.loadtxt
Matft.file.genfromtxt numpy.genfromtxt

  • Operation

    Line 2 is infix (prefix) operator.

Matft Numpy
Matft.add
+		 numpy.add
+
Matft.sub
-		 numpy.sub
-
Matft.div
/		 numpy.div
.
Matft.mul
*		 numpy.multiply
*
Matft.inner
*+		 numpy.inner
n/a
Matft.cross
*^		 numpy.cross
n/a
Matft.matmul
*&		 numpy.matmul
@
Matft.equal
===		 numpy.equal
==
Matft.not_equal
!==		 numpy.not_equal
!=
Matft.less
<		 numpy.less
<
Matft.less_equal
<=		 numpy.less_equal
<=
Matft.greater
>		 numpy.greater
>
Matft.greater_equal
>=		 numpy.greater_equal
>=
Matft.allEqual
==		 numpy.array_equal
n/a
Matft.neg
-		 numpy.negative
-
  • Math function
Matft Numpy
Matft.math.sin numpy.sin
Matft.math.asin numpy.asin
Matft.math.sinh numpy.sinh
Matft.math.asinh numpy.asinh
Matft.math.sin numpy.cos
Matft.math.acos numpy.acos
Matft.math.cosh numpy.cosh
Matft.math.acosh numpy.acosh
Matft.math.tan numpy.tan
Matft.math.atan numpy.atan
Matft.math.tanh numpy.tanh
Matft.math.atanh numpy.atanh
Matft.math.sqrt numpy.sqrt
Matft.math.rsqrt numpy.rsqrt
Matft.math.exp numpy.exp
Matft.math.log numpy.log
Matft.math.log2 numpy.log2
Matft.math.log10 numpy.log10
*Matft.math.ceil numpy.ceil
*Matft.math.floor numpy.floor
*Matft.math.trunc numpy.trunc
*Matft.math.nearest numpy.nearest
*Matft.math.round numpy.round
Matft.math.abs numpy.abs
Matft.math.reciprocal numpy.reciprocal
Matft.math.power numpy.power
Matft.math.square numpy.square
Matft.math.sign numpy.sign
  • Statistics function
Matft Numpy
*Matft.stats.mean *numpy.mean
*Matft.stats.max *numpy.max
*Matft.stats.argmax *numpy.argmax
*Matft.stats.min *numpy.min
*Matft.stats.argmin *numpy.argmin
*Matft.stats.sum *numpy.sum
Matft.stats.maximum numpy.maximum
Matft.stats.minimum numpy.minimum
Matft.stats.sumsqrt n/a
Matft.stats.squaresum n/a
  • Linear algebra
Matft Numpy
Matft.linalg.solve numpy.linalg.solve
Matft.linalg.inv numpy.linalg.inv
Matft.linalg.det numpy.linalg.det
Matft.linalg.eigen numpy.linalg.eig
Matft.linalg.svd numpy.linalg.svd
Matft.linalg.polar_left scipy.linalg.polar
Matft.linalg.polar_right scipy.linalg.polar
Matft.linalg.normlp_vec scipy.linalg.norm
Matft.linalg.normfro_mat scipy.linalg.norm
Matft.linalg.normnuc_mat scipy.linalg.norm

About

Numpy-like library in swift. (Multi-dimensional Array, ndarray, matrix and vector library)

Topics

swift image math signal-processing numpy image-processing complex-numbers ndarray matrix-library ndimensional-arrays

Resources

Readme

License

BSD-3-Clause license
Activity

Stars

141 stars

Watchers

1 watching

Forks

22 forks
Report repository

Releases 6

0.3.3 Latest
Jun 10, 2023
+ 5 releases

Packages

No packages published

Contributors 6

Languages