r"""Check the shape validity, and infer the output shape given input shape and operator. Note the both :attr:`shp1`, :attr:`shp2` and the returned shape are feature shapes (i.e. we remove the first dimension, which correspond to graph statistics such as number of nodes, number of edges,
(op, shp1, shp2)
| 8 | |
| 9 | |
| 10 | def infer_broadcast_shape(op, shp1, shp2): |
| 11 | r"""Check the shape validity, and infer the output shape given input shape and operator. |
| 12 | Note the both :attr:`shp1`, :attr:`shp2` and the returned shape are feature |
| 13 | shapes (i.e. we remove the first dimension, which correspond to graph statistics |
| 14 | such as number of nodes, number of edges, etc.). |
| 15 | |
| 16 | We allow applying op on operands with different shapes, according to the |
| 17 | broadcasting semantics of Numpy/Scipy: |
| 18 | https://numpy.org/doc/stable/user/basics.broadcasting.html |
| 19 | |
| 20 | Parameters |
| 21 | ---------- |
| 22 | op : str |
| 23 | The binary op's name, could be `add`, `sub`, `mul`, `div`, `dot`, `copy_lhs`, `copy_rhs`. |
| 24 | shp1 : tuple[int] |
| 25 | The shape of lhs operand. |
| 26 | shp2 : tuple[int] |
| 27 | The shape of rhs operand. |
| 28 | |
| 29 | Returns |
| 30 | ------- |
| 31 | tuple[int] |
| 32 | shape after broadcasting |
| 33 | """ |
| 34 | pad_shp1, pad_shp2 = shp1, shp2 |
| 35 | if op == "dot": |
| 36 | if shp1[-1] != shp2[-1]: |
| 37 | raise DGLError( |
| 38 | "Dot operator is only available for arrays with the " |
| 39 | "same size on last dimension, but got {} and {}.".format( |
| 40 | shp1, shp2 |
| 41 | ) |
| 42 | ) |
| 43 | if op == "copy_lhs": |
| 44 | return shp1 |
| 45 | if op == "copy_rhs": |
| 46 | return shp2 |
| 47 | # operands are padded to have the same dimensionality with leading 1's. |
| 48 | if len(shp1) > len(shp2): |
| 49 | pad_shp2 = (1,) * (len(shp1) - len(shp2)) + shp2 |
| 50 | elif len(shp1) < len(shp2): |
| 51 | pad_shp1 = (1,) * (len(shp2) - len(shp1)) + shp1 |
| 52 | for d1, d2 in zip(pad_shp1, pad_shp2): |
| 53 | if d1 != d2 and d1 != 1 and d2 != 1: |
| 54 | raise DGLError( |
| 55 | "Feature shapes {} and {} are not valid for broadcasting.".format( |
| 56 | shp1, shp2 |
| 57 | ) |
| 58 | ) |
| 59 | rst = tuple(max(d1, d2) for d1, d2 in zip(pad_shp1, pad_shp2)) |
| 60 | return rst[:-1] + (1,) if op == "dot" else rst |
| 61 | |
| 62 | |
| 63 | def to_dgl_nd(x): |
no test coverage detected