[ENH] Feature Constructor: Enable comprehensions and lambdas in expressions

[ales-erjavec]

Issue

Fixes #6265

Description of changes

Enable comprehensions and lambdas in expressions

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #6272 (1d125d8) into master (acb7120) will increase coverage by 0.01%.
The diff coverage is 95.27%.

❗ Current head 1d125d8 differs from pull request most recent head 6e7fc2c. Consider uploading reports for the commit 6e7fc2c to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6272      +/-   ##
==========================================
+ Coverage   86.70%   86.72%   +0.01%     
==========================================
  Files         316      316              
  Lines       68099    68124      +25     
==========================================
+ Hits        59048    59082      +34     
+ Misses       9051     9042       -9     

[janezd]

It works for sum([x ** 2 for x in range(5)]), but sum(x ** 2 for x in range(5)) fails with

  File "/Users/janez/orange3/Orange/data/table.py", line 780, in from_table
    out = array_conv.get_columns(source, row_indices, n_rows,
  File "/Users/janez/orange3/Orange/data/table.py", line 317, in get_columns
    col_array = match_density(_compute_column(col, sourceri))
  File "/Users/janez/orange3/Orange/data/table.py", line 213, in _compute_column
    col = func(*args, **kwargs)
  File "/Users/janez/orange3/Orange/widgets/data/owfeatureconstructor.py", line 1267, in __call__
    return self.__call_table(table)
  File "/Users/janez/orange3/Orange/widgets/data/owfeatureconstructor.py", line 1291, in __call_table
    return np.asarray(y, dtype=self.dtype)
TypeError: float() argument must be a string or a number, not 'generator'

If this is difficult to fix, I'm OK with it - say so and I merge.

[ales-erjavec]

This is a problem with sum definition in the

https://github.com/biolab/orange3/blob/master/Orange/widgets/data/owfeatureconstructor.py#L1214

    ...
    "sum": lambda *args: np.sum(args),

i.e.

sum = lambda *args: np.sum(args)
print(sum(x ** 2 for x in range(5)))
<generator object <genexpr> at 0x1359e5770>

I guess the builtin overrides should be removed?

[janezd]

I guess the builtin overrides should be removed?

Yes.