Add helper functions to take the "head" of vector or matrix objects. #69
Conversation
This uses zero-copy export and import to try to be as efficient (including memory efficient), but it does modify `.gb_obj`, so it's a little funky. I'm thinking of using this to improve the repr for sparse data.
jim22k
left a comment
jim22k
left a comment
There was a problem hiding this comment.
I really like this approach to taking ownership, grabbing a portion of the data, and giving the original data back. Nice trick!
Did you plan to add a method so this can be called as myvec.ss.head(2)?
I'm undecided. I'm tempted to. Probably. What do you think? I may also add I can't guarantee this method for any backend, so it would be inappropriate to add this functionality to |
Also, add `dtype=` keyword argument to head so that it better matches `to_values`.
|
We don't have |
| (mask is None or mask.structure) | ||
| and df.shape != matrix.shape | ||
| and min(matrix._nvals, max_rows if matrix._nvals <= max_rows else min_rows) | ||
| > 2 * df.count().sum() |
There was a problem hiding this comment.
Note to self: this condition was chosen so I wouldn't need to rewrite any tests. "Friendlier" conditions for when to switch to the sparse repr may exist.
| # SS, SuiteSparse-specific: head | ||
| num_rows = matrix._nvals if matrix._nvals <= max_rows else min_rows | ||
| if matrix._is_transposed: | ||
| cols, rows, vals = matrix._matrix.ss.head(num_rows, sort=True) |
There was a problem hiding this comment.
Note to self: sorting here for a better repr result. If we find this is too slow for large objects, we can update it so we only sort for small objects (but we need to figure out where the threshold may be).
2 participants
This uses zero-copy export and import to try to be as efficient (including memory efficient) as possible, but it does modify
.gb_obj, so it's a little funky.I'm thinking of using this to improve the repr for sparse data.
This was pretty quick to pull together. These aren't the most robust tests, but they're better than nothing. Improving the repr is where the "real" work will be!