Builtins frontend #22007
Builtins frontend #22007
Conversation
|
Thanks for contributing to Ivy! 😊👏 |
ivy/functional/frontends/builtins/math/power_and_logarithmic_functions.py
Show resolved
Hide resolved
|
|
||
|
|
||
| def _infer_return_array(x: Iterable) -> Callable: | ||
| # get module path |
There was a problem hiding this comment.
Could you add a docstring that explains what this decorator does overall? It is not very obvious when it should be used.
There was a problem hiding this comment.
This function should always be used within @to_ivy_arrays_and_back decorator to infer the return array type. Let's consider the jax frontend. In this case, if decorator receives jax.Array, its frontend implementation or list the output will always be converted to a frontend implementation of jax.Array. However, in case of builtins the behaviour changes to: almost_any_array -> ivy.Array -> almost_any_array. In other words, we need to track what array to convert back to. So, if our built-in function recieves np.ndarray or its frontend implementation, the output will be converted to np_frontend.ndarray and so on. It also has a default behaviour. By default we convert the output to a frontend implementation of array which corresponds to current backend. This was done to cover the cases, when function accepts scalar arguments, but returns an array. For example, range. But yeah I will add a docstring.
| return x.ivy_array | ||
|
|
||
| # convert native arrays and lists to ivy arrays | ||
| elif isinstance(x, ivy.NativeArray): |
There was a problem hiding this comment.
If we want lists to get converted, we have to change this to else.
I think however that ivy functions can already handle array-likes, in which case you would just need to change the comment, not the condition.
There was a problem hiding this comment.
Yeah, I just forgot to modify the comment.
|
|
||
| jax.config.update("jax_enable_x64", True) | ||
|
|
||
| ret, frontend_array = fn(*args, **kwargs) |
There was a problem hiding this comment.
The way inputs_to_ivy_arrays and outputs_to_frontend_arrays have been designed right now they can't be used individually. They would only work when used together in to_ivy_arrays_and_back.
E.g. ret, frontend_array = fn(*args, **kwargs) would cause an error if fn was an unwrapped frontend function because there would not be a frontend_array returned.
…pes in outputs_to_frontend_arrays
|
|
||
|
|
||
| def to_ivy_arrays_and_back(fn: Callable) -> Callable: | ||
| return outputs_to_frontend_arrays(inputs_to_ivy_arrays(fn)) |
There was a problem hiding this comment.
- Make another wrapper inside here like:
@functools.wraps(fn)
def _to_ivy_arrays_and_back(*args, **kwargs):
- Move the
frontend_array = _infer_return_array(args[0]) if len(args) != 0 else Nonethere. - Call
inputs_to_ivy_arrays, and then calloutputs_to_frontend_arrayspassingfrontend_array. - You will need to add
frontend_arrayas a new argument likeoutputs_to_frontend_arrays(fn: Callable, frontend_array: str).
There was a problem hiding this comment.
Thanks! I'll try this approach!
No description provided.