Adding multiple Statistical functions to paddle backend #10869
Conversation
|
hy |
KevinUnadkat
changed the title
| @@ -72,7 +72,7 @@ def std( | |||
| keepdims: bool = False, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.std(x, dim=axis, unbiased=False, keepdim=keepdims) | |||
There was a problem hiding this comment.
please review the input arguments and make sure that the output adheres to them
| @@ -84,7 +84,7 @@ def sum( | |||
| keepdims: bool = False, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return torch.sum(input=x, dim=axis, dtype=dtype, keepdim=keepdims) | |||
There was a problem hiding this comment.
use paddle implementation, and check that the args and kwargs are correct
| @@ -96,7 +96,7 @@ def var( | |||
| keepdims: bool = False, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.var(x, dim=axis, unbiased=False, keepdim=keepdims) | |||
There was a problem hiding this comment.
same comments as above
|
@MahmoudAshraf97 |
| @@ -72,7 +72,7 @@ def std( | |||
| keepdims: bool = False, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.std(x, axis=axis, unbiased=True, keepdim=keepdims) | |||
There was a problem hiding this comment.
you need to implement the correction parameter, the unbiased parameter in paddle sets correction = 1, but it needs to be a float, check the torch backend implementation as it's the closest one to paddle
| @@ -96,7 +96,7 @@ def var( | |||
| keepdims: bool = False, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.var(x, axis, unbiased=true, keepdim=keepdims) | |||
There was a problem hiding this comment.
same comment as std
| @@ -112,7 +112,7 @@ def cumprod( | |||
| dtype: Optional[paddle.dtype] = None, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.cumprod(x, dim=None, dtype=None, name=None) | |||
There was a problem hiding this comment.
you need to implement the exclusive and reverse arguments here
| @@ -124,12 +124,12 @@ def cumsum( | |||
| dtype: Optional[paddle.dtype] = None, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| return paddle.cumsum(x, axis=axis, dtype=dtype) | |||
There was a problem hiding this comment.
same comment as cumprod
|
|
||
|
|
||
| def einsum( | ||
| equation: str, | ||
| *operands: paddle.Tensor, | ||
| out: Optional[paddle.Tensor] = None, | ||
| ) -> paddle.Tensor: | ||
| raise IvyNotImplementedException() | ||
| return paddle.einsum(equation, *operands) |
There was a problem hiding this comment.
Please verify the results with another backend to make sure the implementation is correct
|
I recommend you to read both Ivy documentation and paddle documentation when implementing a function in order to know the difference in implementation because some functions work differently, also check other backends implementations for reference |
| for a in axis: | ||
| size *= x.shape[a] | ||
| if size - correction <= 0: | ||
| ret = paddle.var(x, axis==axis, unbiased=False, keepdim=keepdims) |
| ret = paddle.var(x, axis==axis, unbiased=False, keepdim=keepdims) | ||
| ret = ivy.full(ret.shape, float("nan"), dtype=ret.dtype) | ||
| return ret | ||
| ret = torch.mul( |
There was a problem hiding this comment.
please correct this
| @@ -112,7 +151,28 @@ def cumprod( | |||
| dtype: Optional[paddle.dtype] = None, | |||
| out: Optional[paddle.Tensor] = None, | |||
| ) -> paddle.Tensor: | |||
| raise IvyNotImplementedException() | |||
| dtype = ivy.as_native_dtype(dtype) | |||
There was a problem hiding this comment.
please rewrite this function from scratch and ensure that it's free from syntax errors and arguments are passed correctly, or remove it from the PR
Added prod from Statistical to paddle backend