In lists and structs, nested floats are compared via exact equality. Instead, we should correctly use atol and rtol to compare these floats.
For 1-dimensional arrays, we could solve it like this:
elif isinstance(dtype_left, pl.Array) and isinstance(dtype_right, pl.Array):
# For two arrays, we necessarily need to have the same shape. If that isn't
# the case, we cannot observe equality
if dtype_left.shape != dtype_right.shape:
return pl.repeat(pl.lit(False), pl.len())
if len(dtype_left.shape) > 1:
pass # Multi-dimensional arrays not supported yet, but easy in practice
breakpoint()
return pl.all_horizontal(
[
_compare_columns(
col_left=col_left.arr.get(i),
col_right=col_right.arr.get(i),
dtype_left=dtype_left.inner,
dtype_right=dtype_right.inner,
abs_tol=abs_tol,
rel_tol=rel_tol,
abs_tol_temporal=abs_tol_temporal,
)
for i in range(dtype_left.shape[0])
]
)The same idea (at least naively) should extend to lists. We would first compute the length of the largest list, and then iterate with list.get(idx, null_on_oob=True). This isn't great if there are many short lists and few long lists, but it could be a start.
In lists and structs, nested floats are compared via exact equality. Instead, we should correctly use atol and rtol to compare these floats.
For 1-dimensional arrays, we could solve it like this:
The same idea (at least naively) should extend to lists. We would first compute the length of the largest list, and then iterate with list.get(idx, null_on_oob=True). This isn't great if there are many short lists and few long lists, but it could be a start.