2. Currently available features

Numscrypt currently supports:

  • ns_settings.optim_space setting added, default is False. Setting it to True is DISADVISED, since it will result in slow code.

  • ndarray with
    • dtype int32, float32 and float64, shape, views using offset and strides
    • multi-dimensional indexing
    • multi-dimensional slicing
    • reshape
    • astype
    • tolist
    • __repr__ and __str__
    • transpose
    • overloaded operators: * / + - @ (no mixing of ndarray and scalar expressions yet)
  • empty, array, copy

  • hsplit, vsplit

  • hstack.vstack

  • zeros, ones, identity

  • linalg with
    • matrix inversion

3. Systematic code examples: a guided tour of Numscrypt

One ready-to-run code example is worth more than ten lengthy descriptions. The autotest and demo suite, that is part of the distribution, is a collection of sourcecode fragments called testlets. These testlets are used for automated regression testing of Numscrypt against NumPy. Since they systematically cover all the library constructs, they are also very effective as a learning tool. The testlets are arranged alphabetically by subject.

Autotest: Numcrypt autotest demo suite
import org.transcrypt.autotester

import basics
import a_linalg

autoTester = org.transcrypt.autotester.AutoTester ()

autoTester.run (basics, 'basics')
autoTester.run (a_linalg, 'a_linalg')

autoTester.done ()

3.1. Basics: creating and using arrays

Testlet: basics
from org.transcrypt.stubs.browser import *
from org.transcrypt.stubs.browser import __main__, __envir__, __pragma__

# Imports for Transcrypt, resolved run time
if __envir__.executor_name == __envir__.transpiler_name:
    import numscrypt as num

# Imports for CPython, resolved compile time
__pragma__ ('skip')
import numpy as num
__pragma__ ('noskip')

def run (autoTester):
    z = num.zeros ((4, 3, 2), 'int32')
    autoTester.check ('Zeros', z.tolist (), '<br>')
    
    o = num.ones ((1, 2, 3))
    autoTester.check ('Ones', o.astype ('int32') .tolist ())
    
    i = num.identity (3, 'int32')
    autoTester.check ('Identity', i.tolist (), '<br>')
    
    # shape: 2 blocks x 3 rows x 4 columns
    a = num.array ([
        [
            [1, 1, 2, 3],
            [4, 5, 6, 7],
            [8, 9, 10, 12]
        ], [
            [100, 101, 102, 103],
            [104, 105, 106, 107],
            [108, 109, 110, 112]
        ]
    ])
    
    # print (a)
    
    autoTester.check ('Matrix a', a.tolist (), '<br>')
    autoTester.check ('Transpose of a', a.transpose () .tolist (), '<br>')
    
    b = num.array ([
        [
            [2, 2, 4, 6],
            [8, 10, 12, 14],
            [16, 18, 20, 24]
        ], [
            [200, 202, 204, 206],
            [208, 210, 212, 214],
            [216, 218, 220, 224]
        ]
    ])
    
    bp =  b.transpose ((2, 1, 0))
    
    autoTester.check ('Matrix b', b.tolist (), '<br>')
    autoTester.check ('Permutation of b', bp.tolist (), '<br>')
        
    c = num.array ([
        [1, 2, 3, 4],
        [5, 6, 7, 8],
        [9, 10, 11, 12],
    ], 'int32')
    
    autoTester.check ('Shape strides c', tuple (c.shape), tuple (c.strides), '<br>')
    autoTester.check ('Matrix c', c.tolist (), '<br>')
    
    ct = c.transpose ()
    autoTester.check ('Shape strids ct', tuple (ct.shape), tuple (ct.strides), '<br>')
    autoTester.check ('Transpose of c', ct .tolist (), '<br>')

    cs0, cs1 = num.hsplit (c, 2)
    autoTester.check ('Matrix cs0', cs0.tolist (), '<br>')
    autoTester.check ('Matrix cs1', cs1.tolist (), '<br>')

    ci = num.hstack ((cs1, cs0))
    autoTester.check ('Matrix ci', ci.tolist (), '<br>')
    
    cts0, cts1, cts2 = num.hsplit (ct, 3)
    autoTester.check ('Matrix cts0', cts0.tolist (), '<br>')
    autoTester.check ('Matrix cts1', cts1.tolist (), '<br>')
    autoTester.check ('Matrix cts2', cts2.tolist (), '<br>')

    cti = num.hstack ((cts2, cts1, cts0))
    autoTester.check ('Matrix ci', cti.tolist (), '<br>')
    
    d = num.array ([
        [13, 14],
        [15, 16],
        [17, 18],
        [19, 20]
    ], 'int32')
    
    autoTester.check ('Matrix d', d.tolist (), '<br>')
    dt = d.transpose ()
    autoTester.check ('Permutation of d', dt.tolist (), '<br>')
    
    ds0, ds1, ds2, ds3 = num.vsplit (d, 4)
    autoTester.check ('Matrix ds0', ds0.tolist (), '<br>')
    autoTester.check ('Matrix ds1', ds1.tolist (), '<br>')
    autoTester.check ('Matrix ds2', ds2.tolist (), '<br>')
    autoTester.check ('Matrix ds3', ds3.tolist (), '<br>')

    di = num.vstack ((ds3, ds2, ds1, ds0))
    autoTester.check ('Matrix di', di.tolist (), '<br>')
    
    dts0, dts1 = num.vsplit (dt, 2)
    autoTester.check ('Matrix dts0', dts0.tolist (), '<br>')
    autoTester.check ('Matrix dts1', dts1.tolist (), '<br>')

    dti = num.vstack ((dts1, dts0))
    autoTester.check ('Matrix dti', dti.tolist (), '<br>')
    
    v0 = num.array (range (10)) 
    v1 = num.array ((1, 2, 3, 1, 2, 3, 1, 2, 3, 1))
    
    __pragma__ ('opov')
    a [1, 0, 2] = 77777
    el = b [1, 2, 3]

    bsl0 = b [1, 1 : 3, : ]
    bsl1 = b [1 : 2, 1 : 3, :]
    bsl2 = b [1, 1, :]
    bsl3 = b [1, 1 : 3, 1]
    bsl4 = b [ : , 1, 1]
    bsl5 = b [1, 1 : 3, :]
    bsl6 = b [1, 1 : 3, 1 : 4]
    bsl7 = b [1, 2 : 3, 2 : 4]
    
    bpsl0 = bp [1, 1 : 3, : ]
    bpsl1 = bp [1 : 2, 1 : 3, :]
    bpsl2 = bp [1, 1, :]
    bpsl3 = bp [1, 1 : 3, 1]
    bpsl4 = bp [ : , 1, 1]
    bpsl5 = bp [3, 1 : 3, :]
    bpsl6 = bp [2 : 4, 1 : 3, 0 : 1]
    bpsl7 = bp [2 : 4, 2 : 3, 1 : 2]
    
    sum = a + b
    dif = a - b
    prod = a * b
    quot = a / b
    dot = c @ d
    vsum = v0 + v1
    vel = vsum [6]
    vsum [6] = 70
    __pragma__ ('noopov')
    
    autoTester.check ('El a [1, 2, 3] alt', a.tolist (), '<br>')
    autoTester.check ('El b [1, 2, 3]', el, '<br>')
    
    autoTester.check ('Sl b0', bsl0.tolist (), '<br>')
    autoTester.check ('Sl b1', bsl1.tolist (), '<br>')
    autoTester.check ('Sl b2', bsl2.tolist (), '<br>')
    autoTester.check ('Sl b3', bsl3.tolist (), '<br>')
    autoTester.check ('Sl b4', bsl4.tolist (), '<br>')
    autoTester.check ('Sl b5', bsl5.tolist (), '<br>')
    autoTester.check ('Sl b6', bsl6.tolist (), '<br>')
    autoTester.check ('Sl b7', bsl7.tolist (), '<br>')
    
    autoTester.check ('Sl bp0', bpsl0.tolist (), '<br>')
    autoTester.check ('Sl bp1', bpsl1.tolist (), '<br>')
    autoTester.check ('Sl bp2', bpsl2.tolist (), '<br>')
    autoTester.check ('Sl bp3', bpsl3.tolist (), '<br>')
    autoTester.check ('Sl bp4', bpsl4.tolist (), '<br>')
    autoTester.check ('Sl bp5', bpsl5.tolist (), '<br>')
    autoTester.check ('Sl bp6', bpsl6.tolist (), '<br>')
    autoTester.check ('Sl bp7', bpsl7.tolist (), '<br>')
    
    autoTester.check ('Matrix sum', sum.tolist (), '<br>')
    autoTester.check ('Matrix difference', dif.tolist (), '<br>')
    autoTester.check ('Matrix product', prod.tolist (), '<br>')
    autoTester.check ('Matrix quotient', quot.tolist (), '<br>')
    autoTester.check ('Matrix dotproduct', dot.tolist (), '<br>')
    
    autoTester.check ('Vector', v0.tolist (), '<br>')
    autoTester.check ('Vector', v1.tolist (), '<br>')
    autoTester.check ('El sum old', vel, '<br>')
    autoTester.check ('Vector sum new', vsum.tolist (), '<br>')
    

3.2. Linalg: matrix inversion

Testlet: a_linalg
from org.transcrypt.stubs.browser import *
from org.transcrypt.stubs.browser import __main__, __envir__, __pragma__

# Imports for Transcrypt, resolved run time
if __envir__.executor_name == __envir__.transpiler_name:
    import numscrypt as num
    import numscrypt.linalg as linalg

# Imports for CPython, resolved compile time
__pragma__ ('skip')
import numpy as num
import numpy.linalg as linalg
__pragma__ ('noskip')

def run (autoTester):
    a = num.array ([
        [0, -2, -1], 
        [2, 1, 3], 
        [1, 1, 2]
    ])
    
    autoTester.check ('Matrix a', a.astype ('int32') .tolist (), '<br>')
    
    ai = linalg.inv (a)
    
    autoTester.check ('Matrix ai', ai.astype ('int32') .tolist (), '<br>')
    
    __pragma__ ('opov')
    id = a @ ai
    __pragma__ ('noopov')
    
    autoTester.check ('a @ ai', id.astype ('int32') .tolist (), '<br>')

4. Some more examples: interactive tests

4.1. ns_settings.optimize_space

For time critical operations like @ and inv, slicing operations are avoided. For ‘@’ this happens by copying arrays to ‘natural stride order’. Setting ns_settings.optimize_space to True will avoid this copying to save memory space. In general this is DISADVISED, since it will considerably slow down execution of the @ operator, which is O (n^3).

Benchmark: slicing_optimization
from org.transcrypt.stubs.browser import *
from org.transcrypt.stubs.browser import __pragma__

import numscrypt as ns
import numscrypt.random as random
import numscrypt.linalg as linalg

result = ''

for optim_space in (False, True):
    ns.ns_settings.optim_space = optim_space

    for transpose in (False, True):
        a = random.rand (30, 30)
        
        timeStartTranspose = __new__ (Date ())
        if transpose:
            a = a.transpose ()

        timeStartInv = __new__ (Date ())
        ai = linalg.inv (a)
        
        timeStartMul = __new__ (Date ()) 
        __pragma__ ('opov')
        id = a @ ai
        __pragma__ ('noopov')
        
        timeEnd = __new__ (Date ())
        
        result +=  '''<pre>
Optimized for space instead of time: {}
    
{}: a @ ai [0:5, 0:5] =

{}

Transpose took: {} ms
Inverse took: {} ms
Product took: {} ms
            </pre>'''.format (
            optim_space,
            'natural' if a.ns_natural else 'unnatural',
            str (ns.round (id [0:5, 0:5], 2)) .replace (' ', '\t'),
            timeStartInv - timeStartTranspose,
            timeStartMul - timeStartInv,
            timeEnd - timeStartMul
        )

document.getElementById ('result') .innerHTML = result