add transpose method to numpy frontend

[martinoywa]

This PR addresses Issue #4586

[martinoywa]

Please double-check the documentation for numpy.transpose at https://numpy.org/doc/stable/reference/generated/numpy.transpose.html

Since numpy doesn't normally accept an out keyword argument, I would omit it from the frontend function and pass out=None to the ivy function, to prevent inconsistent behaviour with different backends.

You also need to add hypothesis tests to https://github.com/martinoywa/ivy/blob/master/ivy_tests/test_ivy/test_frontends/test_numpy/test_matrix/test_methods.py

Hi, thanks for the hint on out. Still working on the tests.

[martinoywa]

Do you need help with the hypothesis tests?

Hi, yes. I noticed some changes in the implementation with the addition of classes. I've pushed my implementation based on the changes. It passes locally. Please check.

[martinoywa]

Hi @Nightcrab, I'm having some trouble coming up with hypothesis tests for the method. Would you mind lending a hand?

[Nightcrab]

@martinoywa You might want to use the arrays_and_axes helper function to generate the inputs for this test.

[martinoywa]

That's actually what I was using to generate the test cases but they kept on crashing. This is what I have for the test.

@handle_cmd_line_args
@given(
    x_axes=helpers.arrays_and_axes(
        min_num_dims=1,
        max_num_dims=2,
        min_dim_size=1,
        max_dim_size=2,
        num=1
    ),
)
def test_numpy_transpose(
    x_axes,
    as_variable,
    native_array,
    fw,
):
    x, axes = x_axes
    helpers.test_frontend_method(
        input_dtypes_init=ivy.int32,
        as_variable_flags_init=as_variable,
        num_positional_args_init=0,
        native_array_flags_init=native_array,
        all_as_kwargs_np_init={
            "data": x[0],
            },
        input_dtypes_method=[],
        as_variable_flags_method=as_variable,
        num_positional_args_method=0,
        native_array_flags_method=native_array,
        all_as_kwargs_np_method={
            "axes":axes,
        },
        fw=fw,
        frontend="numpy",
        class_name="matrix",
        method_name="transpose",
    )

[martinoywa]

The error I get is as below. Which I believe is related to the axes and how permute_dims expects it to be, but I'm still struggling to figure out how to fix it.

test_methods.py:119: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../../helpers/function_testing.py:1222: in test_frontend_method
    ret, ret_np_flat = get_ret_and_flattened_np_array(
../../../helpers/function_testing.py:1722: in get_ret_and_flattened_np_array
    ret = fn(*args, **kwargs)
../../../../../ivy/functional/frontends/numpy/matrix/methods.py:122: in transpose
    return ivy.permute_dims(self.A, axes=axes, out=None)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = (ivy.array([[0]]),), kwargs = {'axes': [None], 'out': None}

    @functools.wraps(fn)
    def new_fn(*args, **kwargs):
        """
        Catch all exceptions and raise them in IvyException
    
        Parameters
        ----------
        args
            The arguments to be passed to the function.
    
        kwargs
            The keyword arguments to be passed to the function.
    
        Returns
        -------
            The return of the function, or raise IvyException if error is thrown.
        """
        try:
            return fn(*args, **kwargs)
        except (IndexError, ValueError) as e:
            raise ivy.exceptions.IvyError(fn.__name__, str(e))
        except Exception as e:
>           raise ivy.exceptions.IvyBackendException(fn.__name__, str(e))
E           ivy.exceptions.IvyBackendException: numpy: permute_dims: 'NoneType' object cannot be interpreted as an integer

[Nightcrab]

Looks to me like a problem with the frontend function. numpy.matrix.transpose accepts axes=None while ivy.permute_dims doesn't, so you need to implement the default behaviour, which according to numpy docs is this

None or no argument: reverses the order of the axes.

I imagine you can do this by passing the integers n to 0 to ivy.permute_dims where n is the number of dimensions of the input array.

[martinoywa]

Hi, was wondering what you meant by n to 0 to ivy.permute_dims?

[Nightcrab]

If you call ivy.permute_dims(x,[5,4,3,2,1,0]) where x has 6 dimensions for example, what you get is a reversal of the axes (a.k.a a transposition in 6 dimensions).

[martinoywa]

I get it now, thanks.

[martinoywa]

Hi, I coded up what you mentioned earlier but I'm getting this new error which seems to be coming from the helper function that implements arrays_and_axes where it generates an axes that's a mismatch with the arrays. Any ideas on how to solve this?

test_methods.py:118: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../../helpers/function_testing.py:1222: in test_frontend_method
    ret, ret_np_flat = get_ret_and_flattened_np_array(
../../../helpers/function_testing.py:1722: in get_ret_and_flattened_np_array
    ret = fn(*args, **kwargs)
../../../../../ivy/functional/frontends/numpy/matrix/methods.py:125: in transpose
    return ivy.permute_dims(self.A, axes=axes, out=None)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

args = (ivy.array([[0]]),), kwargs = {'axes': [0], 'out': None}

    @functools.wraps(fn)
    def new_fn(*args, **kwargs):
        """
        Catch all exceptions and raise them in IvyException
    
        Parameters
        ----------
        args
            The arguments to be passed to the function.
    
        kwargs
            The keyword arguments to be passed to the function.
    
        Returns
        -------
            The return of the function, or raise IvyException if error is thrown.
        """
        try:
            return fn(*args, **kwargs)
        except (IndexError, ValueError) as e:
>           raise ivy.exceptions.IvyError(fn.__name__, str(e))
E           ivy.exceptions.IvyError: numpy: permute_dims: axes don't match array

[martinoywa]

The transpose function looks as such;

def transpose(self, axes=None):
        if None in axes:
            return ivy.permute_dims(self.A, axes=tuple(range(self.A.ndim)[::-1]))
        return ivy.permute_dims(self.A, axes=axes, out=None)

[Nightcrab]

It seems like arrays_and_axes will not really do what we're looking for which is generating a permutation of size ndims.

I don't know of a helper that does this, so what I would suggest is using dtype_and_values to get the array then write a permutation generator in the test.

You could perhaps use helpers.floats to generate a float between 0 and 1 then use it to seed random.shuffle(y) where y=[x for x in range(array.ndim)].

[ivy-seed]

This PR has been labelled as stale because it has been inactive for more than 7 days. If you would like to continue working on this PR, then please add another comment or this PR will be closed in 7 days.

[martinoywa]

WIP

[martinoywa]

It seems like arrays_and_axes will not really do what we're looking for which is generating a permutation of size ndims.

I don't know of a helper that does this, so what I would suggest is using dtype_and_values to get the array then write a permutation generator in the test.

You could perhaps use helpers.floats to generate a float between 0 and 1 then use it to seed random.shuffle(y) where y=[x for x in range(array.ndim)].

I'm quite unfamiliar with hypothesis testing. How will this generated array help in testing the function?

[Nightcrab]

In general there are helpers to randomly generate test cases, but here it would be best to write your own. With the randomly shuffled array, you can provide test inputs that match what the function is expecting (permutations).

[martinoywa]

So something similar to regular unit tests?

[Nightcrab]

Regular unit tests aren't randomly generated, and you're still using the hypothesis framework. You'd just be writing your own test strategy.

[ivy-seed]

This PR has been labelled as stale because it has been inactive for more than 7 days. If you would like to continue working on this PR, then please add another comment or this PR will be closed in 7 days.

[ivy-seed]

This PR has been closed because it has been marked as stale for more than 7 days with no activity.