Add 2d constructor #208
Add 2d constructor #208
Conversation
| @@ -96,6 +96,20 @@ def dataframe_from_dict(data: Mapping[str, Column[Any]]) -> DataFrame: | |||
| """ | |||
| ... | |||
|
|
|||
| def dataframe_from_2d_array(array: Any) -> DataFrame: | |||
There was a problem hiding this comment.
This should have *, names: list[str] I think, with the length of the list matching the number of columns?
Also a dtype keyword, because we don't really want to get in the business of inferring it (going from array.dtype to a dtype in this namespace is nontrivial; much easier for the caller who has the array namespace at hand).
| """ | ||
| Construct DataFrame from 2D array. | ||
|
|
||
| See `column_from_sequece` for related 1D function. |
There was a problem hiding this comment.
More importantly, I foresee a column_from_1d_array request, because a 1-D array is not a sequence.
There was a problem hiding this comment.
I wasn't aware, thanks
just for my understanding, what makes it not a sequence? from the docs https://docs.python.org/3/glossary.html#term-sequence :
An iterable which supports efficient element access using integer indices via the getitem() special method and defines a len() method that returns the length of the sequence
, don't arrays meet the requirement?
There was a problem hiding this comment.
See data-apis/array-api#481 for len(). A second issue is that iterating over it is going to return 0-D arrays of the dtype of the input array, not Python scalars.
There was a problem hiding this comment.
Are arrays iterable per the array API? I would strongly prefer to differentiate array handling vs iterable handling since you can often handle array zero copy or at least in very efficient ways relative to iterables.
There was a problem hiding this comment.
sure, I've added column from 1d too
| dtypes : Mapping[str, DType] | ||
| Dtype of each column. Must be the same length as ``array.shape[1]``. |
There was a problem hiding this comment.
What type of casting do we allow vs disallow here? I.E. if someone passes an array of type float32 and a mapping that uses bool dtypes here, is that something we expect to support?
There was a problem hiding this comment.
No cross-kind casting? So, float32 to bool would be disallowed?
There was a problem hiding this comment.
Do we allow downcasting within kind or only upcasting? I.e. is float64 array --> float32 column allowed? Or only float32 array --> float32 column?
There was a problem hiding this comment.
is there any array-api precedent to go off of? (haven't looked to carefully yet)
There was a problem hiding this comment.
There is. Here are the casting rules: https://data-apis.org/array-api/latest/API_specification/type_promotion.html#type-promotion. In summary:
- cross-kind casting is undefined, given that not all libraries allow that
- downcasting is disallow in general, and if you're sure that that is what you want, there is a single function (
astype) to explicitly downcast
There was a problem hiding this comment.
thanks - OK to just go with that?
|
merging then, thanks for discussions and review! |
Also a request from scikit-learn