[ENH] Feature constructor optimization

[ales-erjavec]

Issue

Ref: #5911

Parallel (or alternative) to #5949

Description of changes

Avoid iteration over table rows, checks in inner loops, construction of Value instances.

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #5975 (92b2864) into master (369f2c3) will increase coverage by 0.02%.
The diff coverage is 98.90%.

@@            Coverage Diff             @@
##           master    #5975      +/-   ##
==========================================
+ Coverage   86.40%   86.43%   +0.02%     
==========================================
  Files         315      315              
  Lines       67155    67218      +63     
==========================================
+ Hits        58025    58097      +72     
+ Misses       9130     9121       -9     

[ales-erjavec]

Benchmarked using

import statistics

import timeit
import numpy as np

from Orange.data import ContinuousVariable, Domain, Table
from Orange.widgets.data.owfeatureconstructor import ContinuousDescriptor, \
    construct_variables

N = 100_000
table = new_domain = None


def setup():
    global table, new_domain
    var_names = ['Var' + str(i) for i in range(100)]
    variables = [ContinuousVariable(name=var_name) for var_name in var_names]
    domain = Domain(variables)
    test_data = np.random.rand(N, len(variables))
    table = Table.from_numpy(domain, test_data)

    expr = 'Var0 + Var1'
    desc = ContinuousDescriptor(expr, expr, number_of_decimals=None)
    constr_vars = construct_variables([desc], table)
    new_domain = Domain(table.domain.attributes + tuple(constr_vars))


def run():
    table.transform(new_domain)


res = timeit.repeat(
    "run()", setup="setup()", globals=globals(), repeat=10, number=1
)
print(f"Avg: {statistics.mean(res):.4f}, Min: {min(res):.4f}, Std: {statistics.stdev(res):.4f}")

Before: Avg: 0.9430, Min: 0.9309, Std: 0.0086
After: Avg: 0.0451, Min: 0.0447, Std: 0.0002

[markotoplak]

Wow, 20x speedup!