Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/opt_einsum/backends/object_arrays.py: 17%

24 statements  

« prev     ^ index     » next       coverage.py v7.4.0, created at 2024-01-03 07:57 +0000

1""" 

2Functions for performing contractions with array elements which are objects. 

3""" 

4 

5import numpy as np 

6import functools 

7import operator 

8 

9 

10def object_einsum(eq, *arrays): 

11 """A ``einsum`` implementation for ``numpy`` arrays with object dtype. 

12 The loop is performed in python, meaning the objects themselves need 

13 only to implement ``__mul__`` and ``__add__`` for the contraction to be 

14 computed. This may be useful when, for example, computing expressions of 

15 tensors with symbolic elements, but note it will be very slow when compared 

16 to ``numpy.einsum`` and numeric data types! 

17 

18 Parameters 

19 ---------- 

20 eq : str 

21 The contraction string, should specify output. 

22 arrays : sequence of arrays 

23 These can be any indexable arrays as long as addition and 

24 multiplication is defined on the elements. 

25 

26 Returns 

27 ------- 

28 out : numpy.ndarray 

29 The output tensor, with ``dtype=object``. 

30 """ 

31 

32 # when called by ``opt_einsum`` we will always be given a full eq 

33 lhs, output = eq.split('->') 

34 inputs = lhs.split(',') 

35 

36 sizes = {} 

37 for term, array in zip(inputs, arrays): 

38 for k, d in zip(term, array.shape): 

39 sizes[k] = d 

40 

41 out_size = tuple(sizes[k] for k in output) 

42 out = np.empty(out_size, dtype=object) 

43 

44 inner = tuple(k for k in sizes if k not in output) 

45 inner_size = tuple(sizes[k] for k in inner) 

46 

47 for coo_o in np.ndindex(*out_size): 

48 

49 coord = dict(zip(output, coo_o)) 

50 

51 def gen_inner_sum(): 

52 for coo_i in np.ndindex(*inner_size): 

53 coord.update(dict(zip(inner, coo_i))) 

54 locs = (tuple(coord[k] for k in term) for term in inputs) 

55 elements = (array[loc] for array, loc in zip(arrays, locs)) 

56 yield functools.reduce(operator.mul, elements) 

57 

58 out[coo_o] = functools.reduce(operator.add, gen_inner_sum()) 

59 

60 return out