Add normalize_token dispatch for cudf objects#10398
Closed
Add normalize_token dispatch for cudf objects#10398
normalize_token dispatch for cudf objects#10398Conversation
3 tasks
Member
Author
|
I'm going to close this in favor of rapidsai/cudf#13692, since I expect that we will probably want to iterate on the exact logic in cudf/dask_cudf. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
As discussed in dask/dask-expr#80,
tokenizedoes not return a deterministic result for cudf objects. In order for dask-expr to "work" with cudf-backed data, we need the samecudf.DataFrameobject to always return the same token.This PR currently leverages the
hash_valuesmethod to avoid moving all of the cudf data to pandas for tokenization. I cannot think of a "proper" way to definenormalize_tokensuch that we don't need to move data from device to host. One possibility is to tokenizestr(obj._data._data), which returns something like:{'A': <cudf.core.column.numerical.NumericalColumn object at 0x7f12880c17c0> [ 0.8953784288214444, 0.08755770229686265, 0.34997287209485844, -0.3932070666454323, -0.48936633276402525, -0.27516736628435173, 0.345745552676493, -0.8052518333730919, -0.724573970398181, 1.417117733900153, ... 1.964413734663649, 1.4461948823430957, 0.3597774372003223, -1.8124063537988735, 1.3471233020721483, 0.06658665344959998, -2.407065477016253, -0.08984276519450277, 0.0563293241722997, -0.3651402969564298 ] dtype: float64, 'B': <cudf.core.column.numerical.NumericalColumn object at 0x7f12880c1740> [ -0.6808073065312824, 0.7509298536770876, 1.0138677714674298, 0.9697560553365036, -0.6810526127554865, 2.2993594506803046, 1.0055211903308015, 1.4056699179705736, 0.27700903648754815, -0.6962352402999742, ... 1.504129139928046, -1.07522259421438, -0.3427721142143722, 0.15845171832180857, -0.15321560790625618, 0.13699003853682115, -0.0021965876556498997, -1.0362483192703498, 0.8643568908318128, -0.002322298244040977 ] dtype: float64}This approach doesn't account for every row/value, but does account for the schema and data-buffer locations (maybe this is good enough?).
Perhaps @galipremsagar has some ideas :)